YouTube now adding VP9 option to popular videos [Archive]

Nintendo Maniac 64

12th May 2014, 23:27

2016 UPDATE: Sorry, the download and image links are down. If I ever get around to making a new test upload or something, I'll probably redo them.

It's important to note however that the higher-quality fmt22 AVC encodes no longer exist - they're now identical to the non-HFR 720p DASH AVC encode (except in a non-DASH container).

Here's a list of my quality tests from newest to oldest:
[2015-12-20] Lossless 1080p, 30fps (https://www.youtube.com/watch?v=Dj8wZoM-w1c)
[2015-12-20] Lossless 1080p, 2x speed @ 60fps (https://www.youtube.com/watch?v=NbLnNcb0SbE)
[2015-11-06] Lossless 1080p, 30fps (https://www.youtube.com/watch?v=B7Y186Wh_AE)
[2015-11-06] Lossless 1080p, 2x speed @ 60fps (https://www.youtube.com/watch?v=NXPADDVmXU4)
[2015-07-05] Lossless 1080p, 30fps (https://www.youtube.com/watch?v=BEbIkAOCm6A)
[2015-05-01] 6Mbps AVC 720p, 30fps (https://www.youtube.com/watch?v=DFy6yKkV-Jg)
[2014-05-12] 6Mbps AVC 720p, 30fps (https://www.youtube.com/watch?v=BLxHjixnf3c)
[2013-11-12] 3/6Mbps AVC 720p, Half speed @ 15fps (https://www.youtube.com/watch?v=5YWKAKr9W2w)

2015 Update

I've been doing quite a bit of testing and felt that I should probably link the stuff earlier in the thread.

Videos in question
Source & YouTube download mirror: https://www.mediafire.com/#ssmidj916dmw3
2014: https://www.youtube.com/watch?v=BLxHjixnf3c
2015: https://www.youtube.com/watch?v=DFy6yKkV-Jg

[Source]　MP4 AVC (6169kbps): Screenshot A (http://i.minus.com/ibcUeYgI61xBig.png) ~ Screenshot B (http://i.minus.com/iVwePXc6G4oAc.png) ~ Screenshot C (http://i.minus.com/iilkuIQOechoY.png)

2014

DASH AVC (2112kbps): Screenshot A (http://i.minus.com/iboSw4RSHtp8A1.png) ~ Screenshot B (http://i.minus.com/iiTeCt5kesP36.png) ~ Screenshot C (http://i.minus.com/inacAy8QXo8Ew.png)

fmt22 AVC (2886kbps): Screenshot A (http://i.minus.com/insuN23oU8IV5.png) ~ Screenshot B (http://i.minus.com/iZ0UAYwbA8wpN.png) ~ Screenshot C (http://i.minus.com/ibnMvG7YD3Wcmp.png)

WebM VP9 (2182kbps): Screenshot A (http://i.minus.com/i9BMg79d3ReLj.png) ~ Screenshot B (http://i.minus.com/ibiPJNBePjunbh.png) ~ Screenshot C (http://i.minus.com/iyr6M7rrz8TwM.png)

2015

DASH AVC (2125kbps): Screenshot A (http://i.minus.com/ibk2PLJxjSqlLq.png) ~ Screenshot B (http://i.minus.com/ib0sNe4K5R6u8Z.png) ~ Screenshot C (http://i.minus.com/imgRW8KSOh1ch.png)

fmt22 AVC (2894kbps): Screenshot A (http://i.minus.com/iDg7MKthoJdN.png) ~ Screenshot B (http://i.minus.com/idMGsiSv4VZtJ.png) ~ Screenshot C (http://i.minus.com/igYTokwQOvX7J.png)

WebM VP9 (1686kbps): Screenshot A (http://i.minus.com/iy3GuVxJAlUXc.png) ~ Screenshot B (http://i.minus.com/ibp2LjLZjrLcKU.png) ~ Screenshot C (http://i.minus.com/iQNT1KKBfYOHt.png)

-------------------------------- ORIGINAL POST --------------------------------

Here's a great example of VP9 on YouTube making the video no longer look like crap ~ UPDATE: I mean for 720p, my internet is too slow to stream 1080p:
https://www.youtube.com/watch?v=Pq3QwK3-mdc&t=13m20s

To test h.264 just open the same link in a browser that doesn't yet support VP9.

For your convenience, here are links to Chrome Portable (supports VP9) Firefox ESR portable (doesn't support VP9):

Chrome: http://portableapps.com/apps/internet/google_chrome_portable
Firefox ESR: http://portableapps.com/apps/internet/firefox-portable-esr

EDIT: Due to its alpha state, BeamNG DRIVE is known to have frame rate hiccups. If you want to test the smoothness and motion performance of VP9 I recommend the following video instead:
https://www.youtube.com/watch?v=YNhnkFEgXw0&t=6m30s

While it has an extremely stable frame rate it's nowhere as visually detailed as the BeamNG video. UPDATE: It does however work great for testing audio/video de-syncing since you can see if the engine revs and gear shifting matches up with the on-screen display in the bottom-right.

EDIT 2: You can download the VP9 WebM files (along with the h.264 formats) from YouTube with one of the following Firefox extensions:
http://www.cys-audiovideodownloader.com/
https://addons.mozilla.org/En-us/firefox/addon/youtube-video-and-audio-dow/

mzso

12th May 2014, 23:45

Here's a great example of VP9 on YouTube making the video no longer look like crap:
https://www.youtube.com/watch?v=Pq3QwK3-mdc&t=13m20s

To test h.264 just open the same link in a browser that doesn't yet support VP9.

For your convenience, here are links to Chrome Portable (supports VP9) Firefox ESR portable (doesn't support VP9):

Chrome: http://portableapps.com/apps/internet/google_chrome_portable
Firefox ESR: http://portableapps.com/apps/internet/firefox-portable-esr

On the other hand video hangs for seconds and when it isn't it's still jerky.

Also vp9 has ~50% more bandwidth here. Which is a LOT.

http://abload.de/img/avc-vp9-yt4hjwz.png

Nintendo Maniac 64

13th May 2014, 00:05

Also vp9 has ~50% more bandwidth here. Which is a LOT.
That's only the 1080p version, which honestly never really had the bandwidth it should have had from the beginning.

720p on the other hand has a much more comparable filesize.

720p fmt247: 427MB VP9 + 33MB vorbis = 460MB
720p fmt136: 382MB h.264 + 52MB AAC = 434MB
720p fmt22: 619MB h.264 + AAC

EDIT: I just downloaded the 720p VP9 version and was testing the playback in Chrome and I noticed that my CPU usage was hitting a maximum average of 70%! Now my CPU is decently old (Athlon 64 x2 4800+ Brisbane G2) but that still seems pretty high to me seeing how I can playback 720p h.264 without hardware acceleration on my CPU underclocked to a measly 1GHz.

Of course, it might just be because it's Chrome rather than a dedicated media player application, but when I tried to play back in VLC v2.1.3 I just got tons of macro-blocking errors...
http://i.imgur.com/2ysnScK.png

xooyoozoo

13th May 2014, 02:39

Couple of still shots from the 720p DASH formats.

Static: AVC (http://i.imgur.com/FeARTdZ.png), VP9 (http://i.imgur.com/SAP6gF4.png)
Motion: AVC (http://i.imgur.com/Sy3jsFV.png), VP9 (http://i.imgur.com/FxwATu6.png)

mpv (built on top of ffmpeg-dev) works fine for me.

Edit: I do wonder how much more processing power/wattage is required to make the VP9 version, and how much better the AVC clip would look if given the same freedoms (though I've long since assumed Youtube switch to hybrid or fixed-function AVC coders for speed and lower electricity bills... because they obviously aren't concerned about quality...).

Nintendo Maniac 64

13th May 2014, 04:01

I must ask, what are you using to watch and screen cap those videos? I imagine that most people on here would be using MPC-HC + LAVfilters, but I just end up with a blank screen...

EDIT: Oh, turns out I just needed to update my version of MPC-HC. I'm a derp!

EDIT 2: Ahhh, it was because I had Matroska unchecked for MPC-HC's internal filters. I did that because I had been using Haali Media Splitter, but honestly I'm not even sure if that's really necessary anymore.

EDIT 3: Ok yeah, VP9 playback in MPC-HC (64bit) seems to typically have about 10-15% lower CPU utilization compared to Chrome, but both seem to have the same absolute maximum at around 80-85% CPU usage. I seriously hope VP9 decoding becomes considerably more efficient in the future (I mean, it's a given that it will, but it's not guaranteed that it'll be noticeably more so).

It'd also be nice if I could find a VP9 video of F-Zero GX in Dolphin, that's a game that always looked terrible on YouTube due to a combination of its very high-speed gameplay and crisp graphics that lack smoothing effects like bloom and motion blur.

EDIT 4:
On the other hand video hangs for seconds and when it isn't it's still jerky.
I'm like 99% sure that what you describe is not the fault of VP9 but rather is the game itself due to the alpha state of BeamNG DRIVE. If you want to test the smoothness and motion performance of VP9 I recommend the following video instead:
https://www.youtube.com/watch?v=YNhnkFEgXw0&t=6m30s

While it has an extremely stable frame rate it's nowhere as visually detailed as the BeamNG video. UPDATE: It does however work great for testing audio/video de-syncing since you can see if the engine revs and gear shifting matches up with the on-screen display in the bottom-right.

Also you can download the VP9 WebM files (along with the h.264 formats) from YouTube with one of the following Firefox extensions:
http://www.cys-audiovideodownloader.com/
https://addons.mozilla.org/En-us/firefox/addon/youtube-video-and-audio-dow/

Kurtnoise

13th May 2014, 08:27

Will that be decodable with the default VP9, or will the branch have to merge or something?

Yes, in order to decode such streams, the current branch must be merged with this one...

mzso

13th May 2014, 11:31

I'm like 99% sure that what you describe is not the fault of VP9 but rather is the game itself due to the alpha state of BeamNG DRIVE. If you want to test the smoothness and motion performance of VP9 I recommend the following video instead:
https://www.youtube.com/watch?v=YNhnkFEgXw0&t=6m30s

While it has an extremely stable frame rate it's nowhere as visually detailed as the BeamNG video.

Also you can download the VP9 WebM files (along with the h.264 formats) from YouTube with the following Firefox extension:
https://addons.mozilla.org/en-US/firefox/addon/complete-youtube-saver/

The 1% wins. It play with normal speed in potplayer. So probably shitty HTML5 playback in chrome is the cause.

zerowalker

13th May 2014, 17:02

How do you guys download VP9?
Are you doing it the Manual way, or do you have an extension (Except for the Firefox one, i use Opera, can use Chrome though for youtube as it supports VP9 for playback).

And that file, or example that was given, i must say that this must be a rare case, the size difference is huge, VP9 get's more bitrate than h264 which normally isn't the case.
I would hope that was the case though, but on my other tests, VP9 and H264 are similar more or less, no Clear winner.

EDIT:

I am quite annoyed by the fact that VP9 is Only used for up-to 1080p.
They had VP9 aimed for 4K, so i hoped they would use it to do the other resolutions as well.

For those who don't know, anything beyond 1080p (1920x1200 for example), get's in a category called "original".
This is Only encoded in h264, it has(as far as i know) never been encoded in Webm, be it VP8 or VP9, which i don't really understand, my only conclusion is that it's not standard as 16:9(1080p) is the "Beast" so they ignore it.

Nintendo Maniac 64

13th May 2014, 20:49

How do you guys download VP9?
Are you doing it the Manual way, or do you have an extension (Except for the Firefox one, i use Opera, can use Chrome though for youtube as it supports VP9 for playback).
You could just download the Firefox ESR Portable I linked to and temporarily use that to download the VP9 formats. Since it's portable you won't have to install anything on your PC.

And that file, or example that was given, i must say that this must be a rare case, the size difference is huge, VP9 get's more bitrate than h264 which normally isn't the case.
I would hope that was the case though, but on my other tests, VP9 and H264 are similar more or less, no Clear winner.
Shows how much my internet connection sucks, you're the second person that immediately compares only the 1080p versions. My connection is only 3Mbps (~360KB/s) so I'm already just barely able to stream 720p let alone 1080p. Not only that but I'm not even sure my CPU is even fast enough to decode 1080p VP9.

Therefore pretty much anything 1080p is completely out of the question for me.

zerowalker

13th May 2014, 20:55

You could just download the Firefox ESR Portable I linked to and temporarily use that to download the VP9 formats. Since it's portable you won't have to install anything on your PC.

Shows how much my internet connection sucks

True, could use that. Though for some reason, with that clip, applications only seem to detect 720p, browsers see 1080p though, not sure why.

Well, i have been there, currently i got 100mbit up/down, but when i had "normal" internet, HD was luxury which i could barely watch, and it was "The Thing".

Luckily internet is developing extremely fast, at least here, it's Fiber all over the place, compared to a few years ago.

What CPU do you have, i know VP9 takes a lot of CPU power, but not sure how much, so would be nice to know what can't handle the Decoding (yet).

EDIT:

https://www.youtube.com/watch?v=C4YrrLUXfZE

Here is a video where i am not sure if VP9 is worse or not then MP4, as VP9 does look sharper, but it has A Lot more artifacts around objects, it looks like a cheap jpeg picture at times.
Sadly it's resized to 1080p as vp9 doesn't do the 1600x1200 it was played at.

But comparing to the original recording, it's awful how much get's blown away on youtubes encodings, the original upload is at 3.4mbps and looks heavenly compared to this.

Nintendo Maniac 64

13th May 2014, 21:16

True, could use that. Though for some reason, with that clip, applications only seem to detect 720p, browsers see 1080p though, not sure why.
fmt22, a 720p h.264 format, is a non-DASH format which means the video and audio are in a single file rather than in separate two streams. The Firefox extension I linked to can download DASH formats as well, which is required for the VP9-encoded WebM files.

Well, i have been there, currently i got 100mbit up/down, but when i had "normal" internet, HD was luxury which i could barely watch, and it was "The Thing".

Luckily internet is developing extremely fast, at least here, it's Fiber all over the place, compared to a few years ago.
You wouldn't happen to live in Europe or near a Google Fiber city, would you?

What CPU do you have, i know VP9 takes a lot of CPU power, but not sure how much, so would be nice to know what can't handle the Decoding (yet).
2.5GHz Athlon 64 x2 4800+ Brisbane 65nm G2

(that enough information for you? :P)

-----------------------------------------------------------

They had VP9 aimed for 4K, so i hoped they would use it to do the other resolutions as well.
Actually back at the Google IO when they first presented VP9 their focus was purely on today's resolutions and quality with less bandwidth. (of course they didn't do this either, but I'm sure most of us would take better quality with the same bandwidth instead)

What happened instead is at the end of the presentation during the Q&A there was one guy that asked if this could be used for 4k content and Google said that it essentually supports up to something like 16k, most likely because most open-source formats are not designed with planned obsolescence. (see: Dolby TrueHD and DTS Master Audio both maxing out at 192KHz 24bit, and yet both FLAC and WavPack both compress better and support sampling rates into the MHz)

Also another guy inquired about Google's use of what was arguably the worst HEVC encoder for their h.265 comparison. After that Google started to push a bit more on comparing VP9 against h.265, but never really came out and specified that it was specifically for 4k video. Instead it ended up being the fact that h.265 was targeting 4k video, and since VP9 was competing with h.265 and was 4k capable, a lot of people made the jump that VP9 was targeting 4k video as well.

For those who don't know, anything beyond 1080p (1920x1200 for example), get's in a category called "original".
This is Only encoded in h264, it has(as far as i know) never been encoded in Webm
This isn't quite true anymore, a few years ago "Original" was instead renamed to either 1440p or 2160p depending on the resolution of the video in question (note that even though the format maxes out at 4096x3072 they still label it at 2160p)

zerowalker

13th May 2014, 21:34

You wouldn't happen to live in Europe or near a Google Fiber city, would you?

I live in Europe yes, near a Google Fiber city though, got no idea of that, i live pretty much in the middle of nowhere, it's surprising there is even fiber here, it's not in the City at all.

2.5GHz Athlon 64 x2 4800+ Brisbane 65nm G2

(that enough information for you? :P)

Hmm, i can decode 1080p vp9 with about 15% use. which in is 30% if i had 2 cores.
But i got a 4ghz OC intel, so not really comparable i guess.

But i am guessing it runs at 100% for you and not keeping up?

Actually back at the Google IO when they first presented VP9 their focus was purely on today's resolutions and quality with less bandwidth. (of course they didn't do this either, but I'm sure most of us would take better quality with the same bandwidth instead)

Better Quality, Same Bandwidth, can't agree more on that.

What happened instead is at the end of the presentation during the Q&A there was one guy that asked if this could be used for 4k content and Google said that it essentually supports up to something like 16k, most likely because most open-source formats are not designed with planned obsolescence. (see: Dolby TrueHD and DTS Master Audio both maxing out at 192KHz 24bit, and yet both FLAC and WavPack both compress better and support sampling rates into the MHz)

Also another guy inquired about Google's use of what was arguably the worst HEVC encoder for their h.265 comparison. After that Google started to push a bit more on comparing VP9 against h.265, but never really came out and specified that it was specifically for 4k video. Instead it ended up being the fact that h.265 was targeting 4k video, and since VP9 was competing with h.265 and was 4k capable, a lot of people made the jump that VP9 was targeting 4k video as well.

Open-Source has that wonderful power of evolving into the never-ending:)

Ah, that explains it.
I actually looked at VP9 before 4k and looked at pretty much all presentations, and now that i think about it, the thing that they went on and on about was simply lowering bandwidth, tehy didn't mention Resolution.

But well, HEVC as proven that 4k looks great at low bitrates compared to x264, so they have to use VP9 for that in the end anyway, else things are looking grim;P

This isn't quite true anymore, a few years ago "Original" was instead renamed to either 1440p or 2160p depending on the resolution of the video in question (note that even though the format maxes out at 4096x3072 they still label it at 2160p)

Ah, that's true, find it pretty disturbing, i am so used to the thing that anything above 1080p was in the "Original" category, i found it quite amusing, but as they have expanded their resolutions, it went into the abyss:(

Nintendo Maniac 64

13th May 2014, 21:50

I live in Europe yes, near a Google Fiber city though, got no idea of that, i live pretty much in the middle of nowhere, it's surprising there is even fiber here, it's not in the City at all.
Google Fiber is a US thing where it's practically the only case of fiber internet in the entire US that's being developed and expanded at all, but is currently only in a few big metro areas. Verizon FIOS was an older Fiber connection that's no longer being expanded but it also was really only in larger metro area, but mostly on the very east US coast.

In the US, being in the middle of nowhere (like I am) is pretty much a guarentee for having slow internet no thanks to regional duopolies between cable and phone companies. (contrary to what our overly-patriotic government officials say, the US is not the pinnacle of free markets)

Hmm, i can decode 1080p vp9 with about 15% use. which in is 30% if i had 2 cores.
But i got a 4ghz OC intel, so not really comparable i guess.
Here's a handy chart that should give you an idea of the CPU performance between different architectures:
http://media.bestofmicro.com/I/O/298752/original/overall.png
(Ivy Bridge and Haswell are not listed, but both would be a bit faster than Sandy Bridge)

Note that I can overclock my CPU to 3GHz, but I rarely feel the need to since I normally run my CPU undervolted at 1.1v and at times I even underclock to 1GHz @ 0.775v (which I can use for non-hardware accelerated 720p 30fps h.264)

But i am guessing it runs at 100% for you and not keeping up?
I'm seeing 50-80% on just 720p 30fps VP9, I haven't even tried 1080p yet.

zerowalker

13th May 2014, 22:09

Here's a handy chart that should give you an idea of the CPU performance between different architectures

I got a Clarkdale, and seeing that mine is probably twice as fast, so that explains it.

That being said, if you don't have any use except in this case for a faster CPU, then there really isn't anything to be bothered with, as you yourself stated:)

I'm seeing 50-80% on just 720p 30fps VP9, I haven't even tried 1080p yet.

Well, no matter how you look at it there, 1080p is impossible at the current state, if there isn't some huge difference in performance between resolutions, you can always download a small video and look.

Nintendo Maniac 64

13th May 2014, 22:14

you can always download a small video and look.
I've already downloaded four 720p VP9 videos that are all like 30 minutes. :p Time & length isn't the issue, I just haven't done it yet.

I do want to note that the F1 2013 video works great for testing audio/video de-syncing because you can see if the engine revs and gear shifting matches up with the on-screen display in the bottom-right.

EDIT: Got a 1080p VP9 video downloaded (the F1 2013 video I linked to before) and yeah it totally pegs my CPU at 100% and the audio and video de-syncs. However, if I run it at 1/2 speed my CPU usage is around 45-60%, so it seems like I'm actually right on the edge. I'll try it when overclocked later today, but in the meantime I'll fiddle with some settings that may reduce my CPU utilization.

EDIT 2: Oh I'm SO close! I set MPC-HC (64bit) to the overlay renderer and set the video frame to "Normal size", closed everything else including explorer.exe and the task manager, and the video was able to play for about a minute without any audio-video de-sync, though there were still quite a few frame-drops. To put it into perspective, when I did have the task manager open the CPU usage even dropped to 98% a couple times.

EDIT 3: Did some new tests and I found something interesting. My HTPC has a 2GHz Conroe-based Core 2 Duo and in everything else seems to perform extremely similarly to my 2.5GHz Brisbane...except in this. If I watch a VP9 video through Chrome then the CPU usage is pretty much the same as it is on my Brisbane, but if I watch the video through MPC-HC (64bit) then the CPU usage is cut in half! This means that even 1080p 30fps VP9 is "only" using around 50-60% CPU utilization. My theory is that MPC-HC's VP9 decoder is taking advantage of SSSE3 which is something my Brisbane lacks, and therefore Chrome most likely just isn't using SSSE3. I'll be able to test this theory more-so when I get my sister's soon-to-be old laptop in 2 weeks since it has a 1.6GHz Core Duo (not Core 2 Duo) which also supports SSSE3 but has IPC only on level with my Brisbane.

Also I'd like to quickly mention that overclocking my Brisbane to 3GHz allowed me to playback the F1 2013 video without any audio-video desync at all, but it still had some frame-drops. It turns out that the "Enhanced Video Renderer" was actually smoother and less choppy than the overlay renderer, but this only seems to be the case if your CPU is fast enough to not have any audio/video de-sync.

2015 Update: The original Core Duo did not support SSSE3, only up to SSE3. I'm a derp.

Monarc

14th May 2014, 11:11

Also I'd like to quickly mention that overclocking my Brisbane to 3GHz allowed me to playback the F1 2013 video without any audio-video desync at all, but it still had some frame-drops. It turns out that the "Enhanced Video Renderer" was actually smoother and less choppy than the overlay renderer, but this only seems to be the case if your CPU is fast enough to not have any audio/video de-sync.

Already tried mpv? http://mpv.io/installation/

I have never tried it on windows, but there are win builds. It's a command line player, only minimal gui.

docs: https://github.com/mpv-player/mpv/tree/master/DOCS/man/en

mandarinka

14th May 2014, 11:42

EDIT 3: Did some new tests and I found something interesting. My HTPC has a 2GHz Conroe-based Core 2 Duo and in everything else seems to perform extremely similarly to my 2.5GHz Brisbane...except in this. If I watch a VP9 video through Chrome then the CPU usage is pretty much the same as it is on my Brisbane, but if I watch the video through MPC-HC (64bit) then the CPU usage is cut in half! This means that even 1080p 30fps VP9 is "only" using around 50-60% CPU utilization. My theory is that MPC-HC's VP9 decoder is taking advantage of SSSE3 which is something my Brisbane lacks, and therefore Chrome most likely just isn't using SSSE3. I'll be able to test this theory more-so when I get my sister's soon-to-be old laptop in 2 weeks since it has a 1.6GHz Core Duo (not Core 2 Duo) which also supports SSSE3 but has IPC only on level with my Brisbane.

Yes, SSSE3 is very useful extension and lately, people have starting to write assemly optimizations that require it (same situation in x265 I think). Hence the K8 (but K10 too) chips are out of luck since they only have SSE2 and are forced to run unoptimized C codepath.

But more importantly, all the assembly optimizations in ffmpeg's VP9 decoder are only written for 64bit mode. 32bit mode is IIRC unoptimised, because the author doesn't care about 32bit mode. If your Chrome is 32bit, then it doesn't utilise SIMD, but 64bit MPC does.

Some users may find that ffvp9 is a lot slower than advertised on 32bit; this is correct, most of our SIMD only works on 64bit machines. If you have 32bit software, port it to 64bit. Can’t port it? Ditch it. Nobody owns 32bit x86 hardware anymore these days.

Quite a misconception, but alas, that's how it is. On the other way, who cares about VP9, really. It would be a problem if they acted the same way about HEVC.

Nintendo Maniac 64

14th May 2014, 23:05

On the other way, who cares about VP9, really. It would be a problem if they acted the same way about HEVC.
Come on, let's not turn this into one of those threads...

zerowalker

14th May 2014, 23:11

The point in VP9 in my thought isn't to actually go head to head with HEVC, as in Reality, HEVC will most likely "Win the war".

However, VP9 is something i am looking forward to on Youtube and similar sites, if it can bring better quality overall, even if it uses lower bitrate, i am happy.
And it clearly seems like it will be able to do it in the future, as Youtube's settings on H264 are extreme, while the VP9 settings probably are not as flexible, leaving them at Okay settings.

So in the end, I and probably others, hope for the Internet Videos to increase in Quality, and also Reduce in Size (I don't care about that however).

Nintendo Maniac 64

14th May 2014, 23:27

Already tried mpv? http://mpv.io/installation/

I have never tried it on windows, but there are win builds. It's a command line player, only minimal gui.

I've tried the program before but not with VP9. It doesn't seem to really do any better than MPC-HC; when I'm not overclocked there are less frame-drops but there's more drastic audio/video de-syncing (likely due to the reduced frame-dropping).

foxyshadis

14th May 2014, 23:35

EDIT 3:My theory is that MPC-HC's VP9 decoder is taking advantage of SSSE3 which is something my Brisbane lacks, and therefore Chrome most likely just isn't using SSSE3. I'll be able to test this theory more-so when I get my sister's soon-to-be old laptop in 2 weeks since it has a 1.6GHz Core Duo (not Core 2 Duo) which also supports SSSE3 but has IPC only on level with my Brisbane.

Core Duo has only up to SSE3, no SSSE3 support. That first showed up with Core 2. Most Core Duo laptops are upgradeable to a Core 2 Duo Merom, though, if you have one lying around.

Nintendo Maniac 64

14th May 2014, 23:46

Forgive me for sounding facepalm-worthy, but I am a 1337 haxor! I have no idea how I did it, but YouTube encoded my F-Zero GX test video into VP9 even though my video only has 80 views.

http://www.youtube.com/watch?v=BLxHjixnf3c

Unlike the other two videos I previously linked to, this one is only 4 minutes long so it's quick to download. However, due to the fact that I added in audio after-the-fact, it is not optimal for testing any audio/video de-sync. Also it's only available in 720p, maybe I'll upload 1080p in the future.

You can compare the YouTube encoding to the original uploaded file here:
http://www.mediafire.com/watch/s525ycmc05nbohs/C_-_1280x720_@_30fps.mp4

-------------------------------------

Core Duo has only up to SSE3, no SSSE3 support. That first showed up with Core 2. Most Core Duo laptops are upgradeable to a Core 2 Duo Merom, though, if you have one lying around.
Hmm, I could have sworn that Yonah also supported SSSE3...

Nevertheless Yonah uses Socket M and most Merom CPUs use Socket P, and I have no spare Socket M nor Socket P CPUs.

To put things into perspective, my 2.5GHz Brisbane and my HTPC's 2GHz Merom are the fastest CPUs I own.

zerowalker

14th May 2014, 23:56

Nice, though it may be cause you seem to be rather famous on Youtube (compared to me you are Obama;P), so perhaps that has some influence in the System where it picks it's choices, just a guess though.

My own Videos are non-VP9 except for 3, which have 20-70k views, everything else (have checked pretty many) should be h264 sadly.

Nintendo Maniac 64

15th May 2014, 00:04

I've edited in the download link for the MP4 I originally uploaded to YouTube into my previous post so that you can compare the uploaded file to the YouTube encoding.

-----------------------

My videos only have that many views because the content of my videos are what Nintendo calls "evergreen", that is the videos in question constantly get a steady stream of views that build up over the years which makes a difference because most of my videos are 3 years old or more. Unlike people with lots of YouTube subscribers, my videos get very very few views in the beginning and then start to get more views per day over time as they place higher in search results.

In other words, people search out the content the videos that I just happen to upload rather than follow me and watch my videos because I upload them.

pieter3d

15th May 2014, 00:38

If I watch a VP9 video through Chrome then the CPU usage is pretty much the same as it is on my Brisbane, but if I watch the video through MPC-HC (64bit) then the CPU usage is cut in half!

If the VP9 stream is encoded using more than one tile, the decoder can use multiple threads, multi-core decode. That would explain the CPU usage being cut in half.

xooyoozoo

15th May 2014, 01:13

Forgive me for sounding facepalm-worthy, but I am a 1337 haxor! I have no idea how I did it, but YouTube encoded my F-Zero GX test video into VP9 even though my video only has 80 views.

http://www.youtube.com/watch?v=BLxHjixnf3c

On 2.8ghz IVB-EP with master-git ffmpeg, outputting to null gives

-threads 1
182 fps: webm-247
240 fps: mp4-136

-threads 4
420 fps: webm-247
632 fps: mp4-136

Nintendo Maniac 64

15th May 2014, 02:39

I must say that fmt136 h.264 on my F-Zero GX video looks like utter crap, especially around the 2 minute mark. :p

If the VP9 stream is encoded using more than one tile, the decoder can use multiple threads, multi-core decode. That would explain the CPU usage being cut in half.

It's using two threads in Chrome.

On 2.8ghz IVB-EP with master-git ffmpeg, outputting to null gives

-threads 1
182 fps: webm-247
240 fps: mp4-136

-threads 4
420 fps: webm-247
632 fps: mp4-136

This is in-line with what I'm seeing on my HTPC's Core 2 Duo; decoding VP9 only requires about 50% more CPU utilization than h.264 does on a dual core CPU with SSSE3.

zerowalker

16th May 2014, 07:13

I see, i guess it comes from the Tile decoding, and since they use different kinds, they clearly should differ.
I am guessing x265 is primary using one core, and then using the second if it overflows, instead of using them in parallel.

VP9 my be depending on the Tiles, and i have no idea how these work, but perhaps if one is taking a lot of space (or whatever they do), the assigned core will get loaded, while the other may chill, and this may be going up and down throughout the video.

This pretty much render it useless to compare the CPU load on cores.

Oh also, on Youtube, is there a way to search for VP9 videos?
I know you can search (or at least you could) for Webm videos Only, but sadly everything is Webm now, and that's VP8 at 240/360p, making it meaningless filtering it down.

https://www.youtube.com/watch?v=Ghgbycqb92c

Here is another VP9 clip where it's sick how the bitrate differs, VP9 get's less than half, making it looks much worse (still quite good if you think about the bitrate though).
Don't understand why the bitrate is so Dynamic between clips, i thought they had a VBR at a certain range.

Nintendo Maniac 64

16th May 2014, 09:33

If you want more VP9 videos just find some popular YouTuber channels. I personally have been using the various members of the Mindcrack server - I already linked to kurtjmac and VintageBeef and I also know that Zisteau also has VP9 encodings.

BTW, it would seems that the VP9 encodings don't come in until about 2 to 3 days or so after a video is uploaded, so you'll want to check the recent but not too recent videos, such as those that are a week old. This means a user that regularly uploads videos is more likely to have videos within that date-window, which is again why I've been using Mindcrack members as a test.

If you want a simple list of links to those Mindcrack YouTubers just use the right-sidebar on this subreddit:
http://www.reddit.com/r/mindcrack

zerowalker

16th May 2014, 09:41

The thing is, i would like to find specific content, or rather "optimal content".

For example, most people will upload severely compressed videos.
These, in my view, are extremely bad to compare to, if the original video looks bad, you have no mean to know which one is better, and you can't tell how good it looks.

Now of course this is normal as it's not that common to have Fast enough internet to be able to upload at transparent bitrates.

However in this case of comparing VP9 on Youtube, i would like to have this, so i usually search on Lossless etc, cause it's either that, or my own videos which i know how it was encoded, and most are at CRF-12 and the like.
But sadly the only VP9 videos i got are very old ones, and that was before i got fast internet, and to make thing worse they are downsized, making it a bad choice to compare.

About the 2-3 days wait for VP9 encoding, i think it can differ, but probably 1 day is at the best (perhaps a bit lower or higher), but generally is probably at what you say 2-3 days.
Then it comes to the other ones, which simply seems to go after a certain view, for example, i got an 17k views video at 720p, that is only mp4, but 20k at 1200p is vp9 (but encoded at 1080p).
So my guess is that for "Not so famous?" channels, 20k may be the normal range in most cases.

Nintendo Maniac 64

16th May 2014, 09:58

Well, in terms of having a good quality source, I believe that kurtjmac's 1080p videos are uploaded with a whopping 20Mbps (about half the bitrate of a Blu-ray), so he's probably a safe bet. Also someone like ImAnderZEL is also likely safe since he uploads at 1440p and lives in Sweden which means crazy fast upload speeds for easy uploading of high quality encodes.

zerowalker

16th May 2014, 10:02

Ah, as long as it was encoded normally with x264 at medium, it should look very good indeed.

As for 1440p, i guess you mean, if they can afford the resolution, bitrate should be affordable as well?

Sadly as i said before though, 1440p doesn't exist in VP9 as far as i know, i pray that it will get added though, downsizing is something i just can't stand.

Nintendo Maniac 64

16th May 2014, 10:08

My point is that if the guy is uploading at 1440p then the source quality should be good enough for 1080p.

Also I meant to say that kurtjmac uploads at 20Mbps for his 1080p videos, which isn't quite as uber but is still just half the bitrate of a Blu-ray.

zerowalker

16th May 2014, 10:22

Not really sure what you mean with 1440p being an indicator towards the quality of 1080p.
But well, is his internet is fast, i doubt he care for making them small, at least i don't, if it takes 10min or 1 hour is nothing compared to 5-10 hours i had to wait before for some decent quality.

20mbps is okay, but only if it's done at good quality settings and encoder.

Would be interesting to see how vp9 would fare when it comes to handling the same bitrates as it's mp4 counterpart, cause now it only wins or loss cause of the huge bitrate difference.
Cause if it loss when the bitrate is the same, than that will be a huge disappointment, however i really doubt that it does, at least not overall, perhaps some scenes and stuff like that.

Have tried asking Youtube about vp9 and 1440p some time ago, but sadly they don't seem to answer, not sure if i write at the right place though, would really like to get in contact with some technical staff that could tell me the general idea.

Nintendo Maniac 64

16th May 2014, 11:08

Would be interesting to see how vp9 would fare when it comes to handling the same bitrates as it's mp4 counterpart, cause now it only wins or loss cause of the huge bitrate difference.
My F-Zero GX video has nearly identical filesize at 720p for DASH h.264 and VP9, and the VP9 version looks miles better. With the video streams by themselves the VP9 version is only a mere 3% larger in filesize (2.1MB).

zerowalker

16th May 2014, 11:11

Really?

Great, that is what i want to see:)
Could you provide a comparison screenshot and shot it here?
Would be nice to see, and would be great for viewers of this forum who checks for simple "updates".

Nintendo Maniac 64

16th May 2014, 11:32

Personally I thought the difference was large enough that the videos would speak for themselves. Therefore you can surely guess for yourself which screenshots are which video format without me telling you. :p

Videos in question
Source & YouTube download mirror: https://www.mediafire.com/#ssmidj916dmw3
2014: https://www.youtube.com/watch?v=BLxHjixnf3c
2015: https://www.youtube.com/watch?v=DFy6yKkV-Jg

[Source]　MP4 AVC (6169kbps): Screenshot A (http://i.minus.com/ibcUeYgI61xBig.png) ~ Screenshot B (http://i.minus.com/iVwePXc6G4oAc.png) ~ Screenshot C (http://i.minus.com/iilkuIQOechoY.png)

2014

DASH AVC (2112kbps): Screenshot A (http://i.minus.com/iboSw4RSHtp8A1.png) ~ Screenshot B (http://i.minus.com/iiTeCt5kesP36.png) ~ Screenshot C (http://i.minus.com/inacAy8QXo8Ew.png)

fmt22 AVC (2886kbps): Screenshot A (http://i.minus.com/insuN23oU8IV5.png) ~ Screenshot B (http://i.minus.com/iZ0UAYwbA8wpN.png) ~ Screenshot C (http://i.minus.com/ibnMvG7YD3Wcmp.png)

WebM VP9 (2182kbps): Screenshot A (http://i.minus.com/i9BMg79d3ReLj.png) ~ Screenshot B (http://i.minus.com/ibiPJNBePjunbh.png) ~ Screenshot C (http://i.minus.com/iyr6M7rrz8TwM.png)

2015

DASH AVC (2125kbps): Screenshot A (http://i.minus.com/ibk2PLJxjSqlLq.png) ~ Screenshot B (http://i.minus.com/ib0sNe4K5R6u8Z.png) ~ Screenshot C (http://i.minus.com/imgRW8KSOh1ch.png)

fmt22 AVC (2894kbps): Screenshot A (http://i.minus.com/iDg7MKthoJdN.png) ~ Screenshot B (http://i.minus.com/idMGsiSv4VZtJ.png) ~ Screenshot C (http://i.minus.com/igYTokwQOvX7J.png)

WebM VP9 (1686kbps): Screenshot A (http://i.minus.com/iy3GuVxJAlUXc.png) ~ Screenshot B (http://i.minus.com/ibp2LjLZjrLcKU.png) ~ Screenshot C (http://i.minus.com/iQNT1KKBfYOHt.png)

zerowalker

16th May 2014, 21:28

Are they really the same bitrate?
The difference is beyond comparable, it looks like 1k bitrate vs 3-4k;S

Just a guess, but is the encoding below 2k?
Cause can't belive it looking like that at 2k+ for mp4.

I am guessing it looks like that all the time?
Cause i know Youtube encodes with h264 have a tendency of being blurry at time.

For example, everything can look sharp and clear (for it's bitrate) and you are all happy.
Then, mostly cause by a slight movement, everything will turn blurry for a while (even if they are standing still right after), and then it will go back.

I find it extremely annoying, and pretty sure it's a "bug?" in the encoder, which i really hope VP9 won't have at it's settings (and from my tests, it doesn't).

foxyshadis

16th May 2014, 22:44

zerowalker

16th May 2014, 22:51

Can't deny this, and i am pretty much crossing my fingers they keep it this way.

As, no matter what there intentions are, we earn on it, and it will increase the force of development on VP9.

Win/Win situation;P

Nintendo Maniac 64

17th May 2014, 00:00

It's amazing that they apparently run x264 on ultrafast/realtime for everything and give it wretched quality, then likely spend hundreds of times as much cpu time on their VP9 encodes that aren't universally compatible. You'd almost think it was intentional, maximizing the differences to sell their format. (The way On2 used to.)
Actually for the YouTubes this method makes a lot of sense. Get the h.264 versions encoded as fast as possible so that the video is rendered ASAP while bringing in the high quality formats later. Heck this is what YouTube does even within h.264 and VP9 - they encode the smallest resolution first and then encode progressively larger resolutions.

Anyway, here's something interesting I noticed. For resolutions 480p and smaller, VP9 actually is of a considerably smaller filesize than the h.264 DASH formats; it would seem that Google is indeed trying to use VP9 to reduce bandwidth. However it would seem that at the same time they're trying to make their HD formats still look good, which can make sense since if the user is watching those resolutions then they probably care at least a bit about the video quality.

zerowalker

17th May 2014, 10:07

I think they pretty much encode small resolutions first cause it's faster, HD takes ages compared to SD, Youtube probably just want Quantity before Quality (Speed in this case).

Hmm, how would you say VP9 looks at those bitrates compared to h264?
And how big difference are we talking, 1000 vs 200?

Really hope you are correct about there aim on HD, i have my doubts as it's extremely dynamic in bitrate however, but as long as VP9 improves, hopefully the bitrate difference won't be large enough to make the efficient difference lose.

xooyoozoo

22nd May 2014, 09:46

mzso

22nd May 2014, 10:16

Not sure when they flipped the switch, but Youtube now has 4K VP9 (format 272). Example 1 (https://www.youtube.com/watch?v=iwwg1EI20_E). Example 2 (https://www.youtube.com/watch?v=Y-Pa8yYXews).

You can grab the original upload for the 2nd example the downloads section here (https://vimeo.com/84171958).

Any tools that support downloading 1080p+ videos? What I have ( 1 (https://addons.mozilla.org/hu/firefox/addon/youtube-video-and-audio-dow/versions/) 2 (https://addons.mozilla.org/hu/firefox/addon/complete-youtube-saver/versions/) ) doesn't show anything for these above 1080p.

foxyshadis

22nd May 2014, 12:45

Any tools that support downloading 1080p+ videos? What I have ( 1 (https://addons.mozilla.org/hu/firefox/addon/youtube-video-and-audio-dow/versions/) 2 (https://addons.mozilla.org/hu/firefox/addon/complete-youtube-saver/versions/) ) doesn't show anything for these above 1080p.

Keepvid (http://keepvid.com/) will show every available stream, including DASH (as separate streams). Just turn Java off, it works much better with Javascript.

mzso

22nd May 2014, 15:04

Keepvid (http://keepvid.com/) will show every available stream, including DASH (as separate streams). Just turn Java off, it works much better with Javascript.

Thanks. I didn't think about keepvid.
(Is it me or did something got bork with the bookmarklet. It redirects me to https which fails to load...)

Nintendo Maniac 64

22nd May 2014, 17:57

If your decoding stacked is built on libav/ffmpeg, and you need faster HEVC decode, you can always use the OpenHEVC fork (https://github.com/OpenHEVC/FFmpeg). Optimizations there eventually percolate outwards, but for whatever reasons, progress is intermittent and OpenHEVC is (still) 2x faster than ffmpeg-master.

Well I was just using MPC-HC v1.7.5 for playing back VP9 and HEVC. Considering that HEVC decoders are ever so slightly older than VP9 decoders I never would have imagined that VP9's decoder in MPC-HC was actually more optimized. This is even more-so the case since, as an outsider (I'm an audio guy), it seems like the video gurus absolutely adore HEVC and yet treat VP9 as the red-headed step child.

Keepvid (http://keepvid.com/) will show every available stream, including DASH (as separate streams). Just turn Java off, it works much better with Javascript.

There's also "Complete YouTube Saver" and "YouTube Video and Audio Downloader" for anyone using Firefox-based browsers:
http://www.cys-audiovideodownloader.com/
https://addons.mozilla.org/En-us/firefox/addon/youtube-video-and-audio-dow/

EDIT: I just realized that YouTube has an actual separate 1440p format, even for h.264. I could have sworn that 1440p and 2160p had the same base format meaning that if your video had 2160p then it didn't have 1440p (and obviously vice versa).

mzso

22nd May 2014, 22:41

There's also "Complete YouTube Saver" for anyone using Firefox-based browsers:
https://addons.mozilla.org/en-US/firefox/addon/complete-youtube-saver/

CYS wasn't updated yet for the new formats.

Nintendo Maniac 64

22nd May 2014, 23:42

CYS wasn't updated yet for the new formats.

Derp, you're right. It's just that CYS already supported VP9 for the other resolutions by the time we found out about it.

zerowalker

22nd May 2014, 23:48

EDIT: I just realized that YouTube has an actual separate 1440p format, even for h.264. I could have sworn that 1440p and 2160p had the same base format meaning that if your video had 2160p then it didn't have 1440p (and obviously vice versa).

I am quite sure of this as well.

Remember the "Original" talk before, i am pretty sure that was the same, being anything above 1080p and up to a certain size.

Then i think they changed name to, 2160p or something, which was anything, 1200p etc, up to 4k or what it was, i think they later changed it to a lower resolution, at least i read something about that.

I guess they added it back cause of 4k being the next "high-end stream", and i guess more resolutions add compatibility, i just hope they could get on with VP9 on at least 1440p, cause my resolution is 1200p which my videos are in most cases.

Nintendo Maniac 64

22nd May 2014, 23:51

It's my understanding that it went like this:

Introduce 4096x3072 as "Original" -> rename format to "2160p" -> introduce a separate "1440p" resolution.

And I just noticed that the Wikipedia article for YouTube (https://en.wikipedia.org/wiki/YouTube) states the following:
The default video stream is encoded in H.264/MPEG-4 AVC format, with stereo AAC audio.[73]

From my tests, if the browser supports both VP9 and MSE via HTML5 then it'll use VP9 as the default; I've seen this with both Chrome and Opera. Can anyone else confirm/deny this?

zerowalker

22nd May 2014, 23:54

That is most probable, it goes hand in hand with what i said i think.

Just not sure, does Youtube support 4k or not?
I mean, 2160 is not 4k, but the naming can be deceiving as they have been before on Youtube.

Also, has any of your new videos become VP9?
Or was it just a rare case back then?

Asmodian

23rd May 2014, 01:41

Isn't 2160p = 4K? I understand 4K = 3840x2160, 8K = 7680x4320. No idea why we switched from vertical to horizontal pixel counts for naming, maybe 2K sounded too 14 years ago? 2160p doesn't sound bad to me but it must have focus tested worse than 4K.

zerowalker

23rd May 2014, 01:54

Oh right, forgot about the different definitions of 4k, 3840 is the correct one, quite frustrated about that, for me 4k is 4000, which means 4096 = 4k in the byte world.

I am guessing the naming sense is simply for the next-gen flash.

For example, it sound much more high end in the market if you say, Skip your 1080p video today and get 4k!
The things that comes to mind is, 1000 *4 = 4k, which is quite the more impact then 2160p which sound like it's just twice as good compared to four times.

Then there is also "Full HD" which is the correct opposie to "4k", i am guessing 2160p is the true 1080p opposite.

From my tests, if the browser supports both VP9 and MSE via HTML5 then it'll use VP9 as the default; I've seen this with both Chrome and Opera. Can anyone else confirm/deny this?

I can confirm that if VP9 is available, it will always be used, even if it doesn't support the resolutions h264 got (as >1080p), which i think can be quite bothersome.
Downsizing has much more impact then the codec difference cause of 2 things.

1: Downsizing is always bad, and is only used for squeezing less bitrates while making it appear better than it is.
2: Higher resolutions on youtube yield higher bitrates, which makes less resolutions with VP9 negligible in quality difference.

Both these together makes it quite clear that the highest resolution will probably always be the winner (In the Youtube case).

Nintendo Maniac 64

23rd May 2014, 02:46

Oh right, forgot about the different definitions of 4k, 3840 is the correct one, quite frustrated about that, for me 4k is 4000, which means 4096 = 4k in the byte world.
Well guess what, YouTube's 4k format supports up to 4096x3072.

Also, has any of your new videos become VP9?
In fact most of my videos have VP9 encodings, the only ones that don't usually only have 1000 or so views anyway. The issue is that my videos with VP9 encodings (other than the F-Zero GX video) were uploaded due to their audio - the actual video stream is just a single still image. :p

However, I do upload those audio-focused videos with completely lossless audio (lossless the entire way - source, editing, and upload) so if you want to do an audio quality comparison then they'd work great.

EDIT: However, if you are doing any audio testing you need to make sure you download the audio via CYS rather than KeepVid - it seems that KeepVid only links to the middle quality DASH audio format while CYS is able to download the low, medium, and high quality DASH audio formats.

EDIT 2: Uh, wow. While doing an audio quality comparison between 1080p WebM and 1080p MP4 I think I may have discovered a semi-widespread vorbis decoding bug. During the very second note of a specific song there's an audible pop that shouldn't be there, but it's there in Foobar2000, Firefox, and Audacity. However, the pop is not there in Chrome, VLC, and MPC-HC. For reference this pop is not present in the AAC encodings as well and to be clear isn't in the original audio either. Also converting the vorbis file to WAV via Foobar makes the pop still be there but converting the same vorbis file via VLC (though at 16 bit) results in the pop being present.

I think I'm going to need to go to the HydrogenAudio guys for this one...

EDIT 3: Even wierder, when I put the save vorbis file into an matroska container and then everything is fine...

Nintendo Maniac 64

23rd May 2014, 23:18

zerowalker

23rd May 2014, 23:38

So then, more on-topic, I just realized that CYS hasn't actually been updated in over 2 months (does that mean YouTube's been doing VP9 since March?!). I've been in contact with the author of that extension over the last several months over several things so I'm going to shoot him an email right now about adding the VP9 4k format.

That's quite nice, over 2 months (i am guessing like 2-4 months) is not that long a time, and still they have pushed out quite a bit of VP9 videos, both new (which is expected) and old, which must be hard if you think of the quantity, even if you take only the 20k views ones.

Though what do you mean VP9 4k format, have i missed something?
As far as i know, VP9 is Only up to 1080p atm.

Nintendo Maniac 64

24th May 2014, 03:38

Though what do you mean VP9 4k format, have i missed something?
You totally did. :p

Not sure when they flipped the switch, but Youtube now has 4K VP9 (format 272). Example 1 (https://www.youtube.com/watch?v=iwwg1EI20_E). Example 2 (https://www.youtube.com/watch?v=Y-Pa8yYXews).

zerowalker

24th May 2014, 03:48

Nice, and it also had 1440p in VP9.

Though, my videos which are 1920x1200/1600x1200 are still only up to VP9, so i guess the trigger is 4K for allowing that middle ground of 1440p.

Still, it's very good that 4k is getting on with VP9, Youtube is pushing it out, some steady development going on here.

mzso

24th May 2014, 11:33

So then, more on-topic, I just realized that CYS hasn't actually been updated in over 2 months (does that mean YouTube's been doing VP9 since March?!). I've been in contact with the author of that extension over the last several months over several things so I'm going to shoot him an email right now about adding the VP9 4k format.

EDIT: Email sent. The guy usually responds anywhere from within a single to to about a week, so it shouldn't take too long.

He told me that he'd have a look. The dev of this extension (https://addons.mozilla.org/hu/firefox/addon/youtube-video-and-audio-dow/versions/) usually works quicker and he already added the formats.

Nintendo Maniac 64

24th May 2014, 23:22

He told me that he'd have a look. The dev of this extension (https://addons.mozilla.org/hu/firefox/addon/youtube-video-and-audio-dow/versions/) usually works quicker and he already added the formats.

Ehh, I much prefer the UI of CYS, feels more "K.I.S.S". Also I particularly like how I can customize what formats it lists and in what order.

Also CYS works with Portable Firefox, though that's partially because I myself emailed the CYS developer about this very thing (CYS will automatically detect ffmpeg.exe if it's in the profile directory).

EDIT: Oh, that extension also misses the high quality DASH audio stream for WebM vorbis, the same one that KeepVid lacks. Considering that I was able to ABX the difference between the two vorbis formats, that's a bit of a deal killer for me.

zerowalker

25th May 2014, 00:47

Finally one of my 720p videos have been made to VP9, must have been this week sometime.

Anyway, this video, as far as i can remember, should have been uploaded losslessly at RGB, so Youtube converted it to YV12 (which reduces compression efficiency a lot on these videos), and then encoded it.

h264 was given nearly 996 kbps, VP9 was given 512kbps.

My review would be that they look nearly the same, just different artifacts.
However, VP9 has some severe ghosting occurring, or well i don't think ghosting is the word.

It appears artifacts in the prediction are of movement, it doesn't occur all the time, but it seems to appear now and then, and it can look quite ugly, and this makes h264 a clear winner.

Here is a picture, it also contains the "ghosting" to show it.

https://docs.google.com/uc?export=download&id=0B_UKJFH8rbiNUW9hcXFCejZ4RE0
https://docs.google.com/uc?export=download&id=0B_UKJFH8rbiNLUJ5YmFGTURhM1k

EDIT:

Oh, that extension also misses the high quality DASH audio stream for WebM vorbis, the same one that KeepVid lacks. Considering that I was able to ABX the difference between the two vorbis formats, that's a bit of a deal killer for me.

Fast OT on this,

Which are you referring to, there is 2 AAC streams, 192kbps VBR and 256kbps CBR (and some worse for lower quality).
Are you saying that Vorbis beats both? (I like Vorbis, but have to say that QAAC is the same in quality, but no idea what Youtube uses, encoders makes a huge difference in quality).

Nintendo Maniac 64

25th May 2014, 03:38

Regarding a lossless video upload, I actually still have the lossless video data for my F-Zero GX video, and it's even 1080p. It's a really big file though (the number 18GB comes to mind) so if I were to upload it I'd likely be limited to 720p even if I did it overnight (my upload bandwidth is only ~75KB/s).

Which are you referring to, there is 2 AAC streams, 192kbps VBR and 256kbps CBR (and some worse for lower quality).
Are you saying that Vorbis beats both? (I like Vorbis, but have to say that QAAC is the same in quality, but no idea what Youtube uses, encoders makes a huge difference in quality).

I'm saying that there's also two Vorbis streams, but you can't access it unless you use CYS. The high quality vorbis is used with 480p and larger while the medium quality is used with 360p and smaller.

zerowalker

25th May 2014, 03:40

I'm saying that there's also two Vorbis streams, but you can't access it unless you use CYS. The high quality vorbis is used with 480p and larger while the medium quality is used with 360p and smaller.

I wanted to know if you think it had better quality compared to the AAC look alike;P

But anyhow, can't drag to much OT.

About the Ghosting, or perhaps it's called "Temporal Distortion?", should i report it as a bug, or is it supposed to behave like that at low bitrates?

Nintendo Maniac 64

25th May 2014, 03:46

About the Ghosting, or perhaps it's called "Temporal Distortion?", should i report it as a bug, or is it supposed to behave like that at low bitrates?

I'm no video guru but I'm pretty sure that's a very normal visual artifact in cases where the image just doesn't have enough bitrate. It's very similar to an overly-compressed JPEG.

I would say that VP9 is just not (yet?) optimized for high-contrast solid colors. Then again, I doubt h.264 would look any better with that bitrate.

I wanted to know if you think it had better quality compared to the AAC look alike;P
I find that YouTube's AAC encoder has more treble but less bass than their vorbis encoder, but I have not (yet) tested how that compares to the original lossless audio source.

zerowalker

25th May 2014, 04:13

The thing is, it's not really the normal artifacts you see.

The artifacts appear to where the next image is.

If you look at the "Green Pig", there is artifacts to the left, and at the next image he goes there, making the artifacts fit into the next image.
So it's like a "artifact leak" from the next image, or at least it seems like it.

https://drive.google.com/folderview?id=0B_UKJFH8rbiNT2pCQ2NzeUJ3cTA&usp=sharing

Here is some images in a sequence, as you can see the artifacts always appear at the direction where the next movement is.
(3rd and 6th images are with the artifacts)

Nintendo Maniac 64

25th May 2014, 04:43

You could always link to the actual video...

zerowalker

25th May 2014, 04:45

Sure thing, https://www.youtube.com/watch?v=gQtufBLB5FI

Thought it would be easier with images to pinpoint the issue;P

Nintendo Maniac 64

25th May 2014, 04:50

Thought it would be easier with images to pinpoint the issue;P

Then use the &t=_m_s URL variable. You can easily access it if you select "share".

And BTW, that "pig" is a moblin - he has a name you know. :p The "Great Moblin".

(I'll admit that I didn't know the boss's name since unlike the polygonal Zelda games they don't announce & formally introduce the boss to you, but I knew it was from one of the Oracle games which allowed me to look it up)

zerowalker

25th May 2014, 04:53

I shared the images, couldn't you access them;S?

And BTW, that "pig" is a moblin - he has a name you know. The "Great Moblin".

haha, i actually knew the name, but i said "pig" to make it understandable if you didn't know the name XD
But yeah, like 90% of the bosses have no name unless you look it up, but that one at least says his own name;P

Nintendo Maniac 64

25th May 2014, 05:06

I shared the images, couldn't you access them;S?

I did, but I still wanted to compare the videos in the separate formats and resolutions and stuff. Also linking directly to the spot in the video helps other people (such as actual video gurus) chime in.

And besides, my father is hogging the internet bandwidth so I can't quickly skip around the video and find that specific spot, so I'll probably have to just download the whole thing...which will have to wait until tomorrow anyway because my father will be hogging the bandwidth for a while tonight.

zerowalker

25th May 2014, 05:09

Okay, well that's good.

For the video, i think it's everywhere actually, best way is when the Camera isn't moving but the sprites are, it's easy to identify then.
So well, you can look pretty much anywhere, it's more obvious on large stuff moving then small.

11:55 is where Great Moblin the "Green Pig" is:)

Nintendo Maniac 64

25th May 2014, 05:13

Then here's a link with a time stamp:
https://www.youtube.com/watch?v=gQtufBLB5FI&t=11m55s

Like I said my father's hogging the bandwidth so I can't download it yet anyway unless I want it to take absolutely forever, so there's the URL in case anyone else is interested in chiming in.

I'm done for now, good night.

zerowalker

25th May 2014, 05:15

Ah forgot you could do links like that, or rather i haven't ever done them myself, so thanks:)

Good night.

Nintendo Maniac 64

25th May 2014, 05:19

Ah forgot you could do links like that, or rather i haven't ever done them myself, so thanks:)
Protip: Use the "Share" button-like-thing below the video on the YouTube page. It includes a built in function to link directly to the spot in the video you're currently viewing.

Ok, NOW good night. (I was still getting ready :p)

EDIT: It's the next day and due to it being Memorial day I unfortunately likely won't have time to look into it today. However, I was able to quickly check the filesize of that F-Zero GX video. That 18GB? Yeah it's more like double that - it's 34GB. It'd be faster for me to stick it on a flash drive and mail the entire drive to someone with Google Fiber.

zerowalker

25th May 2014, 22:18

Protip: Use the "Share" button-like-thing below the video on the YouTube page. It includes a built in function to link directly to the spot in the video you're currently viewing.

Thanks, can't believie i never used that, think i thought it was (Share with your friends, Google+, Facebook etc), will help in occassions:)!

Ok, NOW good night. (I was still getting ready :p)

EDIT: It's the next day and due to it being Memorial day I unfortunately likely won't have time to look into it today. However, I was able to quickly check the filesize of that F-Zero GX video. That 18GB? Yeah it's more like double that - it's 34GB. It'd be faster for me to stick it on a flash drive and mail the entire drive to someone with Google Fiber.

Good Night, you have probably woke by now as the Edit though;P

Ah we had that today, not the exact same thing but pretty similar i guess, but wait which video are we talking about now?

F-Zero GX, is that 34gb?
Or is my video, cause my can't possible be that big, it's like 117mb?

I can download the clip and cut out the part you want, as that will probably be a lot easier, as i got bandwidth to spare, and it makes no use in wasting so much time if you only are looking at a certain part to see an issue.

EDIT:

Okay not even sure how to cut a VP9 file, doesn't work with Virtualdub and others.

Nintendo Maniac 64

26th May 2014, 00:54

F-Zero GX, is that 34gb?

Yes, lossless 1080p 30fps is a whopping 34GB.

zerowalker

26th May 2014, 22:44

Yes, lossless 1080p 30fps is a whopping 34GB.

So the video that you showed before, where VP9 won, was that a lossless upload?

Nintendo Maniac 64

26th May 2014, 23:47

So the video that you showed before, where VP9 won, was that a lossless upload?

No it was not. If you expand the description on the YouTube page you'll find a link to download the MP4 that I uploaded.

zerowalker

27th May 2014, 00:18

No it was not. If you expand the description on the YouTube page you'll find a link to download the MP4 that I uploaded.

Then why do you want the lossless upload;S?

A bit confused, we talked about the VP9 "bug" and the Zelda video, and now it's about a lossless file of another Game.

Was this conversation flow in the same direction, or did i pick up the wrong call;P?

Nintendo Maniac 64

27th May 2014, 23:29

I was merely commenting that I had wanted to upload a lossless version of my own video, but I am unable to due to the huge filesize.

zerowalker

28th May 2014, 22:18

Oh, i was totally out in the woods then.
Sadly i can't be much help, as the only way i could upload it is by letting me into your account (as far as i know?), and that is something i, and surely you don't want.
If however there is a way to upload it without logging into the account, i would gladly lend a hand.

The chance of my uploading being made to VP9 is non-existance as it is except for about 4 cases of over 1.8k videos.

xooyoozoo

9th June 2014, 23:15

I wish there were a way to switch between VP9 and (hardware-accelerated) H264. I'd prefer my battery life to remain unaffected by this codec war.

zerowalker

10th June 2014, 00:33

disable HTML5 player, in other words use browsers which aren't compatible with it (or perhaps there may be ways to force disable them in browsers that do support it?).
Other than that, you are forced to use VP9 if it's available (Even if it hasn't all the resolutions, which i don't like).

vivan

10th June 2014, 03:08

http://www.youtube.com/html5

However some embed videos force using html5 player, so you can't do anything about it.

foxyshadis

10th June 2014, 06:18

Also: MediaSource compatibility table (https://developer.mozilla.org/en-US/docs/Web/API/MediaSource#Browser_compatibility), since that's required for VP9 videos. In FF you can turn it on or off.

zerowalker

12th June 2014, 02:47

Finally!

VP9 is available on 1600x1200 (1440p/Original)!

I could not download it with Firefox so i did it manually, and on this video AVC used 3100 bitrate while VP9 used 3400 (werid that it's actually higher).

VP9 does look better, Not Extremely in most cases, but overall it's much sharper, however it does introduce the "VP9 Artifacts", which is a bit annoying.
However it seems when watching the video it's not that irritating compared to frame-by-frame.

Here is Comparison.

https://docs.google.com/uc?export=download&id=0B_UKJFH8rbiNeXJrUHBVcG5VVW8
https://docs.google.com/uc?export=download&id=0B_UKJFH8rbiNZzR5aTJicXU5X0E

h264 1200p - 3101 kbps (https://docs.google.com/uc?export=download&id=0B_UKJFH8rbiNdHZCWjBMUE5JREE)
VP9 1200p - 3410 kbps (https://docs.google.com/uc?export=download&id=0B_UKJFH8rbiNTER4MWxwUlZ3WlU)

Nintendo Maniac 64

27th June 2014, 11:58

On a somewhat related note, it looks like YouTube is testing 48fps and 60fps support for their DASH formats, including VP9:
https://www.youtube.com/playlist?list=PLbsGxdAPhjv9UrLo19pS8teoRKj7funAy

Note however, VP9 @ 720p 60fps without SSSE3 optimizations and/or support will make your CPU cry (as what happened with my Brisbane :( ).

Interestingly enough, YouTube's flash player seems to drop every-other frame so as to make the resulting framerate 30fps. This means you pretty much need a browser with both HTML5 and MSE support unless you just want to download the video anyway.

Procrastinating

27th June 2014, 14:52

There seems to be a lot of misinformation flying around, but all in all It seems that you cannot get 60fps to work in firefox without dropping or not rendering frames. People generally get it to work in chrome though. Weirdly in firefox I start at 60 and the framerate drops linearly to 15fps..

Nintendo Maniac 64

27th June 2014, 18:44

Yeah, I downloaded the 720p 60fps h.264 MP4 and played it back directly in Firefox and even then it still couldn't handle the 60fps.

Another thing, 1080p VP9 @ 60fps is brutal on the CPU even with SSSE3 optimizations and/or support. My HTPC's 2GHz Conroe-based Core 2 Duo couldn't even decode it in MPC-HC v1.7.5 without huge framedrops unless I enabled D3D fullscreen, and even then the video still slowed down and had the audio cut out at 0:19 in the Battlefield video (though the rest of the video was fine).

EDIT: Uh, is VP9 that much better at 60fps than h.264 or something? The filesizes for the VP9 video data alone is decently smaller than the h.264 video data alone which is moderately surprising considering that the VP9 versions definitely look better.

zerowalker

27th June 2014, 21:14

What do you mean is VP9 better at 60fps vs h264?

Both should be the same when it comes to FPS increase, i think it's pretty linear.

Decoding should be much faster with h264.

Pure quality, VP9 wins in terms of Youtube. If you compare with x264, i am almost certain x264 will win with quite a difference.

Nintendo Maniac 64

28th June 2014, 12:37

I mean that the bitrate of the VP9 60fps videos is considerably lower than the h.264 versions for all resolutions, yet has better quality. From my testing on 30fps videos VP9 didn't have that much less of a bitrate, if at all.

Also, I'll put my lengthy tangent/response here to the following part of your post from the other thread since my reply is more relevant in this thread:
no multi core support (Encoding wise).
Even for a site like YouTube it would still be useful because currently for a 30 minute video you have to wait about 2 days for the VP9 versions to encode, or 3-4 days for an hour-long video. Because of this, there are quite a few people that don't even seem to acknowledge that these VP9 versions even exist because they're watching videos as they come out via their YouTube subscriptions.

You can tell who these people are by how they're complaining that YouTube should be increasing the bitrate before increasing the framerate. It would seem that they apparently missed the memo about YouTube starting to use VP9 and how VP9 uses those pre-existing bits more effectively so as to not leave people with lesser internet connections (like myself) in the dust.

(alternatively, those people are just falling for the audio/video equivalent of the MHz myth where they are looking at the raw bitrate numbers without looking at the actual video image quality itself)

mzso

28th June 2014, 15:26

You can tell who these people are by how they're complaining that YouTube should be increasing the bitrate before increasing the framerate. It would seem that they apparently missed the memo about YouTube starting to use VP9 and how VP9 uses those pre-existing bits more effectively so as to not leave people with lesser internet connections (like myself) in the dust.

(alternatively, those people are just falling for the audio/video equivalent of the MHz myth where they are looking at the raw bitrate numbers without looking at the actual video image quality itself)

Or maybe they just don't like how YT video's quality sucks. Doesn't really matter if it's VP9 or AVC, 60p or 30p. The only good looking videos I saw online were on Vimeo.

Nintendo Maniac 64

28th June 2014, 15:38

Or maybe they just don't like how YT video's quality sucks. Doesn't really matter if it's VP9 or AVC, 60p or 30p. The only good looking videos I saw online were on Vimeo.

You realize that can make it sound like you're proving my very point about the bitrate equivalent of the MHz Myth...

Anyways, I just discovered that it would seem that YouTube's "Auto" resolution function does not factor in playback performance. If you have the bandwidth for a specific resolution but your CPU isn't fast enough, YouTube will still select the higher resolution and you'll end up with frame-drops everywhere unless you manually select a lower resolution setting.

This will be particularly a problem with the 1080p 60fps VP9 stuff, at least until web browsers start incorporating SSSE3 optimizations for VP9 decoding.

EDIT: Yeah, I just tried that 60fps Battlefield video with the 1080p VP9 version on my friend's laptop with a Core i5 520M in Chrome v35 and, again at around the 0:19 mark, it had several frame-drops with the CPU running at near full-tilt (70-90%). Like always, it ran just fine in MPC-HC however with CPU usage around 40%.

mzso

28th June 2014, 16:12

I don't know about you, but that sounds like you're proving my very point about the bitrate equivalent of the MHz Myth.

No you're falling into the denial trap. The way you talk apparently you don't look at either...
Anyway the bitrate is precious little for decent FHD. I don't know about you, but I rarely check the bitrate I just see that the results are poor.

STaRGaZeR

28th June 2014, 16:27

We're on Doom9, I think we can all agree that YouTube videos look like shit no matter the resolution, format, frame rate, etc.

And about VP9, if they're willing to spend 2 days encoding 30 minutes of video, there is not a single technical reason not to use H.264 (x264) with proper settings. It's going to end a lot faster, have more quality on the same bitrate and the possibility of energy efficient decoding via ubiquitous hardware decoders. Too bad the use of VP9 it's not a technical decision at all.

Nintendo Maniac 64

28th June 2014, 16:29

No you're falling into the denial trap. The way you talk apparently you don't look at either...
I re-worded my post before I even noticed that you had replied. The post that you quoted was not quite what I was trying to convey.

Anyway the bitrate is precious little for decent FHD. I don't know about you, but I rarely check the bitrate I just see that the results are poor.
Well consider that I don't even have the bandwidth to stream 1080p. Heck I just barely have the internet bandwidth to stream non-DASH 720p.

EDIT: STaRGaZeR, I'm not ignoring your post, I just don't want to directly respond to what you mention because it's obvious to me that we're on opposite wavelengths. I'd rather this thread not turn into congress.

STaRGaZeR

28th June 2014, 21:26

No problem. But defending that "VP9 uses those bits more effectively" (you said that) is just plain misleading without noting that VP9 is using orders of magnitude more time to encode. Precisely because the reason YouTube's H.264 sucks so much is supposed to be the use of fast settings to reduce encoding time.

zerowalker

28th June 2014, 21:43

In my tests VP9 has been using quite less, and near or a bit more. In other words it's been all over the place.

And for those cases, VP9 lost when it was lower (not sure how much lower, but i think about 50%?).
If they are near, VP9 will win, and higher.. well it's better of course;P

I have no idea how they determine the bitrate. But it must be some kind of CRF. As Pixel Games get extremely low bitrate, so it does detect static images. So it's not just saying (oh 2.5mbs on this one).
Sadly that CRF they are using is quite flawed it seems, so not sure exactly what or how they use it.

As for VP9 at 60fps and it's decoding. It uses about 30-40% of my CPU, nad this is with that Titanfall clip (Which looks crap, guessing it's the original that's bad though as this is 4mbs which should be looking greater than this)

xooyoozoo

28th June 2014, 21:43

They must be using some sort of hybrid ASIC/FPGA/OpenCL/whatever to mass encode all those VP9
videos, right?! Otherwise, they're dumping a bunch of computational wattage on a technologically middling codec that most of youtube's user base will never be able to use with hardware acceleration. It'd be some weird combination of arrogance and belligerence, all for a rather vague political fight.

...at the very least, we better get more favorable MPEG-LA licensing out of this.

zerowalker

28th June 2014, 22:12

Have no idea how they manage to encode it, but it's been concluded that it's for specific sets of videos, often the popular ones, or by popular people. It will also take about 3 days for it to become VP9 (for normal length video).

Hardware Decoding will come, and will most likely be a standard as well as h265 decoding.

Nintendo Maniac 64

28th June 2014, 22:27

VP9 lost when it was lower (not sure how much lower, but i think about 50%?).

Really? I find for the 60fps Battlefield video that the 1080p VP9 version definitely looks better than the 1080p h.264/AVC version even though the VP9 version has lesser bitrate.

that most of youtube's user base will never be able to use with hardware acceleration.
Correction, current user base. Also, contrary to popular belief, h.264 hardware decoding isn't in everything - I own 6 computers that came out since 2006 and only 1 of them has h.264 hardware decoding, and that's only because I had to replace the mobo with a newer one.

zerowalker

28th June 2014, 22:30

Yes, did some tests, while VP9 did look better at some places, i could not lie to myself saying the overall quality was better.
This was with random gameplay videos i found, quite awhile ago.

Things may have changed, so don't take my word for it:P

Nintendo Maniac 64

28th June 2014, 22:35

Well for reference here's the video I'm referring to:
https://www.youtube.com/watch?v=T3ny9zIckP0

zerowalker

28th June 2014, 22:48

What is the Bitrate difference?

The ones i remember looking at was something like, 3.5mbps -> 2mpbs (i think it was something like this?)

Not a small 500kbps change at least.

Nintendo Maniac 64

28th June 2014, 23:11

I am unsure exactly what the bitrate is, but I do know the filesize of the video stream alone (from which you can calculate the average bitrate).

You can see a list of file sizes for all the DASH "video only" streams here:
http://keepvid.com/?url=youtu.be%2FT3ny9zIckP0

zerowalker

28th June 2014, 23:15

45% less bitrate i guess. This is quite a difference. If it really does look overall better, than this is very good.
The tests i did may have been with an older version of VP9 being used, or simply bad videos.

Nintendo Maniac 64

28th June 2014, 23:40

If it really does look overall better, than this is very good.

Can you not download them yourself and see via the very same link above? The video-only streams are playable in both MPC-HC and mpv.

zerowalker

28th June 2014, 23:50

Took a fast look, and h264 seems to win in some places.
But i think VP9 wins in the overall quality difference as h264 makes a blurry mess at certain points.

dapperdan

30th June 2014, 08:28

Decoding should be much faster with h264.

Didn't the ffmpeg devs benchmark equal quality clips and say that VP9 decode was 10% faster?

http://blogs.gnome.org/rbultje/2014/02/22/the-worlds-fastest-vp9-decoder-ffvp9/

In general the differences seemed down to implementation (e.g. the libvpx decoder was about 10% slower than ffh264, and different chipsets varied more or less depending on effort put into them) rather than fundamental differences in complexity.

Nintendo Maniac 64

30th June 2014, 08:47

foxyshadis

30th June 2014, 11:11

That's all single-threaded. Multithreaded VP9 decoding sucks even with ffvp9, and libvpx decoding has near-zero multithreading support just like its encoding, whereas H.264 has very mature support. While raw performance is passable, it means PCs become even bigger space heaters and laptop devices suck down the battery (if they can even decode it, depending on their throttling settings). High bitrate gets much worse in a hurry, because entropy decoding is so slow (like H.264, and one of the problems HEVC addresses). Only the newest mobile devices could handle 1080p VP9 at all at this point, and only at a low bitrate, and that probably won't change until they get hardware support.

The way Google is going about this rollout, I almost wonder if they're trying to force their team to deal with the speed problems that are affecting everyone.

Nintendo Maniac 64

30th June 2014, 17:34

If libvpx hasn't hardly any multithreaded support then why on a CPU without SSSE3 do I see extremely similar CPU usage when decoding a VP9 video in both Chrome 35 and MPC-HC 64bit 1.7.5? (around 60-70% for 1280x720 30fps VP9 on an Athlon 64 x2 4800+ Brisbane G2)

(also for reference, it's not something new - I've had this result for more than 90 days now)

And fun fact: I see more balanced core loading with VP9 via Chrome than with MPC-HC's h.265/HEVC decoder.

Beelzebubu

1st July 2014, 16:37

That's all single-threaded. Multithreaded VP9 decoding sucks even with ffvp9

Wait, what? You have no data to support this statement. The only case where MT sucks is high thread count combined with very low quantizer (or, in reverse, many non-zero coefficients) causing the reconstruction to go down in relative time (or rather, have coefficient parsing go up in relative time) and thus eventually limit the thread speed increase in the default no-mt mode. For typical youtube content, this never happens. Besides, youtube likely uses mt-mode so it's irrelevant anyway and threading gives the same speed increase as any other codec (vp8, h264, you name it).

Beelzebubu

1st July 2014, 16:38

Well, it must be said though that without the SSSE3 optimizations that's in ffvp9, VP9 decoding is like 3x slower than h.264/AVC decoding.

This is important to note because Chrome 35 and Firefox 31 beta seem to lack said SSSE3 optimizations.

Firefox doesn't include ffmpeg for legal reasons. Chrome probably uses libvpx for decoding.

SSSE3 is a runtime-detected optimization path and will be used if your specific computer supports it. This is regardless of what the build host supports or what the minimum platform requirement of the application is.

xooyoozoo

1st July 2014, 23:05

Firefox doesn't include ffmpeg for legal reasons. Chrome probably uses libvpx for decoding.

Oh dang, it looks like you're right about Chrome. Lines 260-266 here (https://git.chromium.org/gitweb/?p=chromium/chromium.git;a=blob;f=media/filters/vpx_video_decoder.cc;h=20416f148077568de3b190e17dd571d1f80ca86f;hb=2c67edd4c43c9470e1dc7960581e89e2409574db) seem to indicate that VP8+Alpha and VP9 are decoded by libvpx instead of ffmpeg.

I thought the occasional slowdowns on my laptop was because Chrome is 32bit in OSX while ffvp9's optimizations only works for 64bit, but it looks like that's not the main issue.

Kurtnoise

2nd July 2014, 09:59

and libvpx decoding has near-zero multithreading support...
This is not true anymore...check out the log (https://chromium.googlesource.com/webm/libvpx/+log/master) for more info.

Nintendo Maniac 64

2nd July 2014, 21:50

Firefox doesn't include ffmpeg for legal reasons. Chrome probably uses libvpx for decoding.

I'm already aware of this. My point was questioning the lack of the multi-threaded performance in libvpx considering that I see near identical performance to ffvp9 when the SSSE3 optimizations aren't present.

BTW, it's officially confirmed that my theory of SSSE3 optimizations was correct, but what I didn't know is that they're essentially part of the 64bit optimizations and is the main reason for advising against decoding with 32bit:
http://lists.ffmpeg.org/pipermail/ffmpeg-devel-irc/2014-July/002201.html

Personally I would have liked if they had recommended using 64bit and an SSSE3-compatible CPU...

foxyshadis

3rd July 2014, 02:25

Wait, what? You have no data to support this statement. The only case where MT sucks is high thread count combined with very low quantizer (or, in reverse, many non-zero coefficients) causing the reconstruction to go down in relative time (or rather, have coefficient parsing go up in relative time) and thus eventually limit the thread speed increase in the default no-mt mode. For typical youtube content, this never happens. Besides, youtube likely uses mt-mode so it's irrelevant anyway and threading gives the same speed increase as any other codec (vp8, h264, you name it).

Last time I tested on the old laptop a couple weeks back (dual Core2 Merom) it just choked and would only use 55-60% cpu at between 10-20fps; ffvp9 handled it great, though. It was a pretty high bitrate since it was a stress test. I don't know if the videos had tiles, but many I've seen don't; do all HD Youtube videos use them?

dapperdan

4th July 2014, 16:55

https://groups.google.com/a/webmproject.org/forum/m/#!msg/webm-discuss/4qFBHpthG3Q/aBsmZtF4V0QJ

That link suggests that all new YouTube VP9 content should have at least 2 tiles, 4 for 1080 upwards, but also that there may still remain older content without tiles.

fumoffu

30th July 2014, 21:43

Is anyone using HTML5 player in Firefox? It's supposed to be default in FF33 (and it is - in Firefox Aurora I don't even see an option to use flash instead) but for now the player is very buggy (problems with buffering and freezing, etc.). At lest for me... Does it maybe work ok for you guys?

mzso

31st July 2014, 00:42

fumoffu

31st July 2014, 02:40

I'm wondering if they can make it work in time. Even in beta youtube should be usable, so they have a month to make it work. I'm worried they will push this feature half done and there will be many angry users...

zerowalker

31st July 2014, 02:54

Well you are better off then me, i use Opera 12, so i am completely stuck.
So i have to use Chrome for Youtube, being Google it works as expected.

dapperdan

31st July 2014, 10:11

Is anyone using HTML5 player in Firefox?

Do you mean specifically with VP9 content?

I've been using the HTML5 player in Firefox for a long time and it's been fine for VP8. But I tried flicking the switch to turn MSE support on, which is required for VP9 on Youtube, and all I got was the audio and a a black rectangle instead of video.

dapperdan

31st July 2014, 10:31

There was a talk live streamed here about Youtube's rollout of VP9:

https://livetalk.tv/vtech29july

I missed it, but there was some talk of the recording being put online at some point too.

BadFrame

1st August 2014, 08:55

But I tried flicking the switch to turn MSE support on, which is required for VP9 on Youtube, and all I got was the audio and a a black rectangle instead of video.

Yeah same here, vp9 is served perfectly in Chrome/Chromium but in Firefox I can't even open a vp9 encoded local file, let alone get vp9 on Youtube, despite Firefox adding vp9 support back in version 28.

sneaker_ger

1st August 2014, 09:24

I can open local VP9 in WebM without problems with FireFox 31.

BadFrame

1st August 2014, 09:46

I can open local VP9 in WebM without problems with FireFox 31.

Doesn't work for me, I get 'Video format or MIME type is not supported', tried both with about:config 'mediasource' enabled and disabled.

I'm on Linux, using Firefox 31

The vp9 videos play perfectly in Chrome/Chromium, mplayer...

sneaker_ger

1st August 2014, 10:04

I'm on Windows 7 x64, "mediasource" is on default (disabled).

MediaInfo or sample?

dapperdan

1st August 2014, 10:08

I've always been able to play basic VP9 files since they started supporting that, but Youtube/Firefox seem to acknowledge that they're not working well together on VP9 content yet, at least partly due to bugs in the new MSE support.

The guy who gave the VP9 at Youtube talk I posted a link to above mentioned this on his twitter feed:

https://twitter.com/_srobertson/status/493637628319657984

"Watching separate implementations of a complex spec (MSE) evolve. Visualizing the operational differences would be revealing. And cool."

Which I guess is a reference to Mozilla's work.

MoSal

1st August 2014, 12:43

Doesn't work for me, I get 'Video format or MIME type is not supported', tried both with about:config 'mediasource' enabled and disabled.

I'm on Linux, using Firefox 31

The vp9 videos play perfectly in Chrome/Chromium, mplayer...

If gstreamer is enabled. Make sure you have the necessary plugins installed.

BadFrame

1st August 2014, 14:50

If gstreamer is enabled. Make sure you have the necessary plugins installed.

Ah, that's most likely the problem, I do not have gstreamer installed.

Do you know if it will be a prerequisite for VP9 on Firefox in the future as well?

VP8 works in Firefox without gstreamer, and VP9 works in Chrome/Chromium, mplayer, VLC without gstreamer.

sneaker_ger

1st August 2014, 15:01

I don't see any hint about VP9 in Firefox not being built-in just like VP8.

fumoffu

1st August 2014, 15:29

VP9 itself works ok. MSE+Vp9 and MSE+H264 doesn't.

MoSal

1st August 2014, 19:33

Ah, that's most likely the problem, I do not have gstreamer installed.

Do you know if it will be a prerequisite for VP9 on Firefox in the future as well?

VP8 works in Firefox without gstreamer, and VP9 works in Chrome/Chromium, mplayer, VLC without gstreamer.

https://bugzilla.mozilla.org/show_bug.cgi?id=763048

If nothing changed since I reported that bug, Firefox won't fall-back to built-in decoders if it was built with gstreamer support. You will need the required gstreamer plugins.

To know how your installed version was built , you can visit about:buildconfig .

BadFrame

1st August 2014, 23:18

https://bugzilla.mozilla.org/show_bug.cgi?id=763048

If nothing changed since I reported that bug, Firefox won't fall-back to built-in decoders if it was built with gstreamer support. You will need the required gstreamer plugins.

Ah, yes this release (from Arch extra repository) was built with '--enable-gstreamer=1.0' so that explains it, here's hoping the bug will some day get fixed so that it will fall back on internal VP9 even if it was built with gstreamer enabled (which I assume is mainly to enable h264 decoding through gstreamer since VP9 is royalty free and shipped 'out of the box' in Firefox ?).

Thanks for the info!

mzso

2nd August 2014, 10:31

https://bugzilla.mozilla.org/show_bug.cgi?id=763048

If nothing changed since I reported that bug, Firefox won't fall-back to built-in decoders if it was built with gstreamer support. You will need the required gstreamer plugins.

To know how your installed version was built , you can visit about:buildconfig .
Another sad bug. A valid issue reported two years ago and then ignored and forgotten.

ceth

29th October 2014, 22:29

Personally I thought the difference was large enough that the videos would speak for themselves. Therefore you can surely guess for yourself which screenshots are which video format without me telling you. :p

Screenshots A: Image 1 (http://i.minus.com/iCxgC72dVYE7R.png) ~ Image 2 (http://i.minus.com/ibjt5PUMqX3n5U.png)

Screenshots B: Image 1 (http://i.minus.com/iIBej6Qrkmhkb.png) ~ Image 2 (http://i.minus.com/ibqxRK86b2dCPI.png)

Jumping late into that thread (read through it all), OMG do want that kind of detail increase for my YT videos ! (but given the low view for my videos I cannot expect VP9 encodes anytime soon).

I don't want to derail that thread too much from the original VP9 topic, but as it turned out a bit into the global question of YT (re)encoding quality, I'd like to share my recent test (and some questions).

I mainly upload PC videogames content (and very racing sim/games oriented). I have just spent hours doing all kind of tests to be able to get the best quality out of YT re-encoding. After finding a first quality increase by exporting lossless from vegas to do the final encoding with another software (I skip on this here), here is what I found so far:

1°) This is certainly well known, doing a slight upscaling let you benefit from a better bitrate a YT will use the profil for the next "step" of resolution they offer (like getting a 1080p YT encode for a video that is just a 720p slightly upscaled). So I upscaled my 1080p source to 1152p, which forced YT to do an additionnal 1440p (tagged as is, but still 1152p, just encoded with higher bitrate).

This let me get a 10Mbps YT re-encode (the "1440p YT version") instead of 4.5Mbps YT was encoding this video for its 1080p YT version. Although unfortunately, despite this 1440p version appears in the list when I download back from YT (I use jdownloader2 (http://jdownloader.org/download/offline) for this, check it ;) ), it does NOT appear as a choice in the YT flash player.

Anyone knows why ? That's strange YT keeps a better 1440p version on their server but does not offer it into their player :confused:
I may do other tests by upscaling more to check if there is a minimum to trigger the fact this better encoding is offer as a resolution choice in their player.

2°) There may be a sweetspot bitrate (for each resolution) concerning the video you upload to YT. My 1080p 8Mbps upload turned to have a significantly better YT re-encoding than the 1080p 10Mbps version I tried. Although I cannot be 100% affirmative on this as I only checked a single image and with a single short sample (20sec), also the appreciation of this "better" quality is partialy subjective. Also I haven't tried to check if another bitrate than 8Mbps for the uploaded video could provide an even better result after YT re-encoding.

Please share any info you would have on this if you tried various bitrate for the same video upload and checked back the result after YT re-encoding.

Here are some screenshots of those tests:

screenshot - 0 (http://img15.hostingpics.net/pics/2460820ORIGINALSHADOWPLAYYUV420.png) Original game capture
screenshot - 1a (http://img15.hostingpics.net/pics/2595821atestx264losslesssharpen02ripbotveryslow1min40solo8Mbps.png) 8Mbps upload
screenshot - 1b (http://img15.hostingpics.net/pics/7660731btestx264losslesssharpen02ripbotveryslow8Mbps1080pretourYT.png) 8Mbps YT 1080p re-encode
screenshot - 1c (http://img15.hostingpics.net/pics/6703961ctestx264losslesssharpen02ripbotveryslow8Mbps1440pretourYT.png) 8Mbps YT 1440p re-encode (resulted in 1152p, as uploaded file)
screenshot - 2a (http://img15.hostingpics.net/pics/8597562atestx264losslesssharpen02ripbotveryslow10Mbps.png) 10Mbps upload
screenshot - 2b (http://img15.hostingpics.net/pics/9457372btestx264losslesssharpen02ripbotveryslow10Mbps1080pretourYT.png) 10Mbps YT 1080p re-encode
screenshot - 2c (http://img15.hostingpics.net/pics/5033502ctestx264losslesssharpen02ripbotveryslow10Mbps1440pretourYT.png) 10Mbps YT 1440p re-encode (resulted in 1152p, as uploaded file)
link to this video on YT (https://www.youtube.com/watch?v=7mH-bYpLfvU&list=PLcnb-N3mVHviDKaldjgAl8Id6Bl-xOChD) (should be locked in germany, and maybe other countries although nothing else than germany was specified. I was experimenting with YT contentID)

-> game capture was made using quicksync @50Mbps VBR (75Mbps peak according MediaInfo) YUV 4:2:0 1072p 30fps (I usually use nvEnc (=shadowplay) but tried quicksync this time. Both offers 50Mbps capture and seem to provide similar capture quality, although I haven't deep checked).

-> rendering chain: capture as described above, using afterburner -> vegas pro 12 -> DNxHD export (almost lossless, only seen a slight difference in 1 grey tone) -> avisynth (Lanczos4Resize(2048,1152) + Sharpen(0.2)) -> x264 lossless -> ripbot264 2pass high@L4.0 very slow tuning:default (=none)

Although this is a bit off topic, as I see several people here uploading gaming content to YT, I would really appreciate if you could share your findings to obtain the best quality after YT re-encoding.

- is it (still) more worth it uploading at 720p instead of 1080p ?
- what triggers the 1440p YT version to appear in their player ? (same could apply to 720p if you want to cheat the same to obtain 1080p YT bitrate for it using a slight upsacle)
- beside popular videos getting VP9 re-encoding (understandable if the goal for YT is to save bandwidth), do you think popular videos (or popular youtubers) also get better quality x264 re-encoding ?
- any other thoughts ?

Although this may seen a waste of time to try to optimize final YT encoding when in the end this work will be spoilt by the awful flash player fluidity (see my recent post on this subject here (http://forum.doom9.org/showpost.php?p=1698043&postcount=7)). But we are stuck with this atm (let hope this will change soon too, it has lasted enough now. The VP9 of the F1 game posted here seemed a bit better fluidity-wise, this is promising too).

We're on Doom9, I think we can all agree that YouTube videos look like shit no matter the resolution, format, frame rate, etc.
If only YT could offer to keep the uploaded version... We are able to produce much better quality at slower bitrate, that's a win-win. Other than compatibility I don't understand why they want to re-encode everything, spoiling the quality of their own offer (videos). I would even be glad to spend a bit more time encoding and uploading a low def version rather than being stuck with an awful YT re-encode.

mzso

30th October 2014, 10:55

This let me get a 10Mbps YT re-encode (the "1440p YT version") instead of 4.5Mbps YT was encoding this video for its 1080p YT version. Although unfortunately, despite this 1440p version appears in the list when I download back from YT (I use jdownloader2 (http://jdownloader.org/download/offline) for this, check it ;) ), it does NOT appear as a choice in the YT flash player.

Anyone knows why ? That's strange YT keeps a better 1440p version on their server but does not offer it into their player :confused:
I may do other tests by upscaling more to check if there is a minimum to trigger the fact this better encoding is offer as a resolution choice in their player.

Interesting. Yours is the first video I noticed this with. Maybe they started doing it recently? Or it's just an error, it's available in the html5 player. I saw something like this backwards where in the HTML5 player 1080p was missing. After I reported it it was fixed after a while. So I guess you should try the "send feedback" link at the bottom of the page.

ceth

30th October 2014, 13:38

Good advice, I'm doing this now.

Nevilne

30th October 2014, 14:22

youtube now supports 60 fps fyi

ceth

30th October 2014, 14:26

Now this is another interesting thing:

I saw you talking about CPU load earlier in the thread. I got another PC in my house that is just meant for web browsing, so this is just an older A64 4000+ single core, overclocked at 2.8Ghz. This PC is daily used to play a popular game in my country, flash based game. However with firefox this game load this CPU at 100% all the time and the result is the game is not smooth. I tested a lot of web browsers and the result was the same everytime, 100% CPU load. All but one: lastest internet explorer. With IE11 this game loads this old single core CPU at only 70%, and the result is the game is smooth as butter ! (IIRC, IE uses its own flash engine, that is more efficient than the adobe one used on other web browsers)

In my previous post I was referring to my recent answer about choppyness of videos read through flash player. When the tests I refer to in this answer were done, I also remembered how IE11 fixed the flash based game CPU load on my weak PC so I gave it a try with YT videos. I had a Q6600@3.4 at this time, and the result of playing YT videos with IE11 was still not smooth, still choppy as with other web browsers.

But as I made this answer here on doom9 about the choppy YT playback, I wanted to give IE11 a try again and reinstalled it. I couldn't test right away as it needed a reboot and I had work going on, but now mzso made his answer about html5 player I tried IE11.

Guess what ? My video (the racing game) is 100% smooth with IE11 ! As smooth as when I do an offline playback with any decent video player. But weird thing: with IE11 this video only offers 2 resolutions choice in the YT player: 340p and 720p... Not only the 1440p is still missing but the 1080p too ! I wondered what player could be used for this IE11 playback and when I right click the video I see this is HTML5 player.

So some questions for you mzso:
- what web browser are you using for your HTML5 player above feedback ?
- what resolutions choice do you get in your HTML5 player for my video ?
- and last, is the video choppy or smooth as butter ? And if not smooth, what is your system config ? (this race video should be smooth 98% of the time, only at some places there is a (heavy) choppyness due to game capture issue where capture dropped to 15fps for a few seconds).

mzso

30th October 2014, 14:47

But as I made this answer here on doom9 about the choppy YT playback, I wanted to give IE11 a try again and reinstalled it. I couldn't test right away as it needed a reboot and I had work going on, but now mzso made his answer about html5 player I tried IE11.

Guess what ? My about video (the racing game) is 100% smooth with IE11 ! As smooth as when I do an offline playback with any decent video player. But weird this: with IE11 this video only offer 2 resolution choice in the YT player: 340p and 720p... Not only the 1440p is still missing but the 1080p too ! I wondered what player could be used and when I right click the video I see this is HTML5 player.

So some questions for you mzso:
- what web browser are you using for your HTML5 player feedback ?
- what resolutions choice do you get in your HTML5 player for my video ?
- and last, is the video choppy or smooth as butter ? And if not smooth, what is your system config ? (this race video should be smooth 98% of the time, only at some places there is a (heavy) choppyness due to game capture issue where capture dropped to 15fps for a few seconds).

I checked HTML5 with chrome, don't use IE and FF's HTML5 player is retarded (no MSE support). Naturally I get all the resolutions up to 1440p. Since in chrome you can't disable the vp9 decoder it would be a miracle if it played smoothly on my aging computer.
If you only get max 720p with IE it's not unsurprising that it has lower CPU usage.

ceth

30th October 2014, 15:03

No, the low CPU usage was about a flash based game. As said above I did compare most popular browsers (incl. chrome) and IE11 is the only one loading the singlecore A64 at a steady 70% where ALL other browsers were resulting in a 100% CPU load. There is no doubt, IE11 is (at least "was", at the time of those test a few monthes ago) the most efficient of all browsers with this flash based game.

And for the video playback smoothness, not related to 720p neither. With FF the 720p version will be as choppy as the 1080p. Even the 144p will still be as choppy. This is not a matter of load but a flash player issue (see my post about this here (http://forum.doom9.org/showpost.php?p=1698043&postcount=7)).

I have not tried yet with HTML5 player, I'm going to test this now with FF and other web browsers too.

ceth

30th October 2014, 17:28

So here are the result of the browsers test I have just done again:

Firefox (v32.0.3)
HTML5: as smooth as offline playback - only 360p & 720p available
Flash: severe choppiness - resolution choice only up to 1080p

IE11 (v11.0.9600.17358IS)
HTML5: as smooth as offline playback - only 360p & 720p available
Flash: -> uses HTML5 despite Flash is selected in YT options

Chrome (v38.0.2125.111 m)
HTML5: almost as choppy as flash... - 240p - 360p - 480p - 720p - 1080p - 1440p available
Flash: severe choppiness - resolution choice only up to 1080p

Opera (v25.0.1614.68)
HTML5: almost as choppy as flash... - 240p - 360p - 480p - 720p - 1080p - 1440p available
Flash: severe choppiness - resolution choice only up to 1080p

Safari (v5.1.7 - 7534.57.2)
Not compatible with YT anymore...

What a mess...
All resolutions available in HTML5 from chrome & opera but their HTML5 playabck is almsost as choppy as the flash player playback...
FF & IE11 have smooth HTML5 playback but YT only show 360p & 720p versions in the player...

But at least HTML5 playback is better than what is was several monthes ago, at least for FF & IE11, as it now offers a perfect smoothness that was not the case when I tested earlier this year. If YT could offer all resolution choices in their HTML5 player for FF & IE and the perfect smoothness was kept with resolution higher than 720p (and I guess it would), there would be no reason to use the flash player anymore. There would FINALLY be a valid solution for perfectly smooth playback for YT videos played directly in the web browser.

I'm going to send that feedback to YT, I hope they will react to it.

mzso

30th October 2014, 19:22

Now I don't know what you mean by choppyness because most people (including me) certainly don't experience it.

ceth

30th October 2014, 19:54

Choppiness means not fluid/smooth playback. You are very likely to be affected as you use chrome. A 30fps video like my race one should be "perfectly" smooth on a 60hz/120hz monitor (I mean as smooth as 30fps can be. Ofc 60fps or more will be even smoother. But when talking about flash or chrome/opera HTML5 choppiness we are talking about VERY unsmooth playback, as is frames were dropped very very often).

mzso

30th October 2014, 19:58

Choppiness means not fluid/smooth playback. You are very likely to be affected as you use chrome. A 30fps video like my race one should be perfeclty smooth on a 60hz/120hz monitor (let aside 60fps or more will be even smoother, ofc. But when talking about flash or chrome/opera HTML5 choppiness we are talking about VERY unsmooth playback, as is frames were dropped very very often).

You can monitor dropped frames in the HTML5 player if you enabled statistics. With 720p I got zero dropped frames after playback stabilized. Didn't have such issues with flash either for the past few years. At least in full screen with active HW acceleration.

ceth

30th October 2014, 20:22

Where did I wrote it would drop frames ?... I said the result is similar to many dropped frames.
And no dropped frames in the statistics doesn't garantee a fluid playback, I have 0 droped frames in chrome and the playback is not smooth at all. On FF & IE11 it is.
I personally never heard someone saying he had smooth flash playback and I never had a fluid flash playback myself on several different PC, HW on or off.
Just do a google search to see how many complains there are about it. If you don't notice it maybe you are not sensitive to smoothness issues and/or you are watching videos where it is less obviously visible. On a racing game video the smoothness issues just become so obvious due to the very frequent horizon straffing when the car turns.

foxyshadis

30th October 2014, 23:33

I have a feeling a large part of the problem has to do with interactions with the video driver. I can report that on a new high-end gaming laptop with nvidia/Intel optimus, Firefox Flash player drops frames like crazy, whereas Firefox and IE11's HTML5 don't. Chrome (HTML5 only) is also perfectly smooth on that system.

It's weird that you only see the non-DASH formats on IE and FF, for me DASH sizes show up. I wonder what causes that.

ceth

31st October 2014, 03:00

You see the 1440p version of my racing video in FF with both the standard YT flash player and its HTML5 player ?
ps: my current system is a win7 x64 desktop with an I7 4770K + nvidia GTX670 with lastest driver.

foxyshadis

31st October 2014, 03:20

Almost; on Flash it only goes up to 1080p, on HTML5 I get 1440p as well. At 1440p my laptop can no longer play it, while 1080p plays fine; on Flash it's a total fail. On my work PC (Intel HD 4000 w/ Flash only) it hits barely 8 fps no matter what size I choose.

ceth

31st October 2014, 03:33

This is weird I only see 360p/720p in both IE11 & FF for HTML5, doesn't make sense, unless YT doesn't offer the same to all countries. I'm going to check this using a proxy, as I don't see what else could cause this difference.
edit: just verified, doesn't seem to be a country lock.

mzso

31st October 2014, 12:00

So apparently 60fps videos are here. Naturally they don't work in firefox, only Chrome and IE... With Chrome I get VP9 (for a lot of videos) which of course is a slideshow at 1080p60...
I wish I could disable the vp9 decoder...
https://www.youtube.com/watch?v=_zPm3SSj6W8
https://www.youtube.com/watch?v=5L562H8a2Fs

Nintendo Maniac 64

5th November 2014, 05:09

Naturally they don't work in firefox, only Chrome and IE... Actually it works in Chrome, IE, Opera, and Safari. For Firefox it works in a Nightly version.

With Chrome I get VP9 (for a lot of videos) which of course is a slideshow at 1080p60...
Try 64bit Chrome if you aren't already using it, it has some performance improvements in the VP9 decoder.

I wish I could disable the vp9 decoder...
Change your user agent to report Windows XP (specifically "Windows NT 5.1"); I just tested it myself and it works to force YouTube to serve h.264.

As was mentioned in the video posted earlier, they don't serve VP9 videos to XP machines since it's assumed that they are older and/or lower-end PCs and therefore don't have the CPU grunt.

mzso

5th November 2014, 13:40

Actually it works in Chrome, IE, Opera, and Safari. For Firefox it works in a Nightly version.

Try 64bit Chrome if you aren't already using it, it has some performance improvements in the VP9 decoder.
Nope. It doesn't work in any FF version, because flash doesn't support it and the html5 player is still dysfunctional in nightly.
I already use 64 bit chrome, but VP9 is just too cpu heavy.

Actually it works in Chrome, IE, Opera, and Safari. For Firefox it works in a Nightly version.
Change your user agent to report Windows XP (specifically "Windows NT 5.1"); I just tested it myself and it works to force YouTube to serve h.264.

As was mentioned in the video posted earlier, they don't serve VP9 videos to XP machines since it's assumed that they are older and/or lower-end PCs and therefore don't have the CPU grunt.

Is UA faking doable in chrome?

Edit:
Well, I tried "Mozilla/5.0 (Windows NT 5.1) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/40.0.2188.2 Safari/537.36" with this (https://chrome.google.com/webstore/detail/user-agent-selector/fnbmdojpgjpmjjmnjdnbobcdhenmmgod/related), but it didn't work I still got VP9.

Edit2:
It works with this one (https://chrome.google.com/webstore/detail/user-agent-switcher/ffhkkpnppgnfaobgihpdblnhmmbodake).

nevcairiel

5th November 2014, 14:11

I already use 64 bit chrome, but VP9 is just too cpu heavy.

Admittedly my CPU is rather fast, but I don't even get 1% CPU usage watching any of the two videos you linked on 1080p60.

Whats this, a wooden PC? :)

mzso

5th November 2014, 14:31

Admittedly my CPU is rather fast, but I don't even get 1% CPU usage watching any of the two videos you linked on 1080p60.

Whats this, a wooden PC? :)

No just an Intel e6750@3,2GHz. Although I still doubt yours is a 100 times faster. :)

Nintendo Maniac 64

6th November 2014, 00:07

Nope. It doesn't work in any FF version, because flash doesn't support it and the html5 player is still dysfunctional in nightly.

Did you enable MSE via about:config?

Admittedly my CPU is rather fast, but I don't even get 1% CPU usage watching any of the two videos you linked on 1080p60.

Whats this, a wooden PC? :)

Uh, I hate to break it to you, but no currently-existing x86 CPU is that fast. Are you absolutely sure that you're running VP9 with 1080p60? (see: "Stats for nerds")

mzso

6th November 2014, 01:48

Did you enable MSE via about:config?

Yes. It hangs around three second in...

Nintendo Maniac 64

6th November 2014, 03:44

Yes. It hangs around three second in...

Well then it technically works, just not well. :p

mzso

15th November 2014, 15:09

zerowalker

15th November 2014, 22:04

Chrome works fine for me. You sue your CPU isn't bottlenecking?
Cause for example 4k VP9 is choppy for me, as it reaches 50% of my CPU (it can only use 2 cores it seems, got 4)

mzso

16th November 2014, 01:26

Chrome works fine for me. You sue your CPU isn't bottlenecking?
Cause for example 4k VP9 is choppy for me, as it reaches 50% of my CPU (it can only use 2 cores it seems, got 4)

No it plays mp4 with the faked user agent.

Nintendo Maniac 64

16th November 2014, 21:56

No it plays mp4 with the faked user agent.

Does your GPU support h.264 decoding? Most GPUs within the last 5 years do, but particularly those that are older may not be able to adaquately handle 60fps HD content.

Try disabling hardware acceleration and/or DXVA in your browser.

mzso

17th November 2014, 02:53

Does your GPU support h.264 decoding? Most GPUs within the last 5 years do, but particularly those that are older may not be able to adaquately handle 60fps HD content.

Try disabling hardware acceleration and/or DXVA in your browser.

My setup can play 1080p 60fps AVC fine with HW acceleration. Not so much without it, not enough CPU power.

Nintendo Maniac 64

18th November 2014, 02:48

My setup can play 1080p 60fps AVC fine with HW acceleration. Not so much without it, not enough CPU power.

Did you try it anyway, particularly with 720p? Not all HW acceleration is created equally (see: VLC vs MPC-HC).

mzso

18th November 2014, 23:21

No difference with AVC. VP9 probably still drops frames because the lack of CPU power. I get around the same CPU usage for it as with 1080p60, but not as flat. The decoder must be really crappy, AVC can max out my cpu.

Nintendo Maniac 64

19th November 2014, 00:04

As crazy as it sounds, apparently IE11's h.264 HW accelerated decoder works well for 60fps videos. Might as well give it a whirl.

mzso

19th November 2014, 00:32

As crazy as it sounds, apparently IE11's h.264 HW accelerated decoder works well for 60fps videos. Might as well give it a whirl.

I have no trouble With HWA or software decoding. But it does look like Firefox (and Chrome) uses a crappy renderer. I actually have IE disable don my computer. :)

Nintendo Maniac 64

20th November 2014, 00:04

Firefox actually has several different renderers which can be seen via about:support and adjusted via about:config - Direct2D (azure), Direct3D (layers), and software (it has less stable OpenGL varients as well).

mzso

20th November 2014, 00:57

Nintendo Maniac 64

4th December 2014, 19:46

VP9/WebM seems to be no longer limited to popular videos and/or popular YouTube channels, case in point:
https://www.youtube.com/watch?v=Y1KJc4kltok
https://www.youtube.com/user/Rush3fan1/videos

Also the amount of time it takes for VP9/WebM versions to appear seems to have been cut down dramatically. For a 30+ minute video, you used to have to wait maybe 3 days, but now it's there in less than 24 hours.

EDIT: From what I can tell, VP9/WebM versions now typically appear somewhere between 19 and 22 hours after a video has been uploaded.

EDIT 2: ...and now I've just come accross two videos that are over 24 hours old but lack VP9/WebM versions even though videos posted only a couple hours earlier have them.

zerowalker

6th December 2014, 12:46

I can tell you that it's kinda limited to popular videos.
I got like 2310 videos. And only a handful (10-30) are VP9.

I noticed that what triggers it for me is that they get about 50-100 views within a 2 days period, then it will be encoded to VP9.
If i get 100 views in a month, i am pretty sure nothing happens, it has to be in a short period of time.

It may also be that i upload so many videos that i am far down the list, the trigger system might be different depending on certain factors.

I can also say that old videos that are 60fps, will become 60fps if they are encoded to VP9. (At least if they got encoded to VP9 after they added it).

Nintendo Maniac 64

6th December 2014, 20:34

Well the videos I'm looking at are into the 5-digits for view count.

But I have a theory - it's based on when the video went live. If it was posted before 3pm on Thursday, then it'll get VP9/WebM within that sub 24-hours time period that I stated. If it's posted after, then it'll be a while longer because that time period starts butting up against the weekend traffic from the other side of the world - remember that only ~30% of YouTube traffic comes from the US.

The thing about Friday afternoon is, I theorize that YouTube's servers see considerably more load starting at Friday afternoon all the way to very early Monday.

EDIT: I just checked on a 3 minute video that has over a million views that was uploaded about 30 hours ago, and sure enough - no VP9/WebM version. For reference, the video uploaded just before it has VP9/WebM even though it has less views and wasn't even 24 hours older.

EDIT 2: Still want to confirm that no video I can find that has been uploaded within the last 48 hours has VP9/WebM versions.

EDIT 3: 60 hours later, still no VP9/WebM...maybe tomorrow morning/afternoon.

EDIT 4: 72 hours hours later, now with 2.5 million views, yet no VP9/WebM. Did something break on YouTube's servers?

EDIT 5: The VP9/WebM versions have finally started appearing. It would seem that, whether intentional or not, no VP9/WebM encoding occurs on the weekend.

EDIT 6: Hmmm... it would seem that encoding VP9/WebM versions are indeed still prioritized for videos with higher view counts. A video with 9k views still doesn't have a VP9/WebM encode even though it's been 120 hours (though again, the video before it has had VP9/WebM since Friday even though it only has 6k views).

EDIT 7: Hmm... I wonder if VP9/WebM versions appear in bulk every 72-96 hours? Because even 1+ million view videos aren't getting VP9/WebM encodes after 48 hours.

EDIT 8: Now 96+ hours, and yet again, still no VP9/WebM... fyi I'm using the following YouTube channels for referencing (all numbers are video view counts):

2M+ - PewDiePie (http://youtube.com/user/pewdiepie/videos)
~300k - EthosLab (http://youtube.com/user/ethoslab/videos)
80k-300k - SethBling (http://youtube.com/user/sethbling/videos)
20k-80k - VintageBeef (http://youtube.com/user/vintagebeef/videos)
30k-50k - ImAnderZEL (http://youtube.com/user/imanderzel/videos)
15-30k - Zisteau (http://youtube.com/user/zisteau/videos)
10k-20k - GuudeBoulderfist (http://youtube.com/user/guudeboulderfist/videos)
5k-15k - KurtJMac (http://youtube.com/user/kurtjmac/videos)
3k-8k - PyroPuncher (http://youtube.com/user/pyropuncher/videos)
1k-2k - SuperMCGamer (http://youtube.com/user/supermcgamer/videos)

De-M-oN

10th December 2014, 22:21

I flew through the text and saw a discussion about browser support of 60fps.

Firefox can do VP9 at HTML5, but you have to enable MSE in about:config first. I dont know how bugfree it works though. I think it has a cause that its not enabled at default.

Flash Player is able to playback 60fps video smooth and with much more complexity than youtube. So I dont know what this shitty youtube player messes up :/ JW Player does it fine @ 2048x1152 resolution.
Try this video: http://killerinstinct.ath.cx:2000/Video/SS-Retro-Enc-DM-60fps.html

But the chrome browser - yeah there works 60fps fine, but there is something which really really annoys me about chrome: It playbacks the videos in pc range, instead of tv range. It does the same mistake like VLC. And this as a result of weak saturated brightened up playback is really ugly :/
And I just dont get it. Google should know that their videos are in tv range? o.O And it doesnt matter if html5 or flash - it does it on both.
Firefox uses tv range. Even Internet Explorer uses tv range.

And if a video gets vp9 seems totally random like a dice. My channel is not that huge, but I get sometimes vp9, sometimes not. Totally random.

mzso

11th December 2014, 18:13

But the chrome browser - yeah there works 60fps fine, but there is something which really really annoys me about chrome: It playbacks the videos in pc range, instead of tv range. It does the same mistake like VLC. And this as a result of weak saturated brightened up playback is really ugly :/
And I just dont get it. Google should know that their videos are in tv range? o.O And it doesnt matter if html5 or flash - it does it on both.
Firefox uses tv range. Even Internet Explorer uses tv range.

It's a driver issue. Change the settings: https://wiki.videolan.org/VSG:Video:Color_nVidia

De-M-oN

11th December 2014, 22:40

oh thx. That helped.

Nintendo Maniac 64

16th December 2014, 03:58

Ok, now this is odd - there have been no new VP9/WebM encodes in over a week now. I'm going to guess that they're tweaking their encoding setup or something.

Nintendo Maniac 64

19th December 2014, 21:24

Forgive the double-posting, but this is really stange - now some VP9/WebM encodes are disappearing. That Explosive Drive video I linked to earlier, citing it as an example of a video with low views that still had VP9/WebM encodes? Yeah, that video no longer has them.

Link again for your convenience:
https://www.youtube.com/watch?v=Y1KJc4kltok

EDIT: Looking around more, it seems like every single video uploaded in the last 30 days have had their VP9 encodes removed, no matter who it was uploaded by.

12 hours later EDIT: ...and now most of them are back, and a bunch of newer videos got VP9 encodes as well.

zerowalker

20th December 2014, 12:55

Hmm, perhaps there was something wrong with the encoder and they removed them just in case. Or it was just some corruption or mistake going on.

mzso

21st December 2014, 16:38

Nintendo Maniac 64

22nd December 2014, 00:08

I don't see webm's for that video. Only for 360p.

That's my point - they were available back when I linked to it on the previous page.

foxyshadis

22nd December 2014, 00:23

Are they different encodes? Maybe they fixed a major bug or reduced the size significantly and wanted to go back and do them again.

Nintendo Maniac 64

22nd December 2014, 00:54

Are they different encodes? Maybe they fixed a major bug or reduced the size significantly and wanted to go back and do them again.

Don't know - I never kept any VP9/WebM videos that were at most a month old.

NikosD

29th January 2015, 10:28

Youtube officially moved to HTML5 and VP9 by default on browsers supported:
http://youtube-eng.blogspot.gr/2015/01/youtube-now-defaults-to-html5_27.html

Patrik G

29th January 2015, 13:29

sneaker_ger

29th January 2015, 13:41

Try Chrome 64 Bit. Not much else you can do about it, I think.

mzso

29th January 2015, 17:30

VP9 not so great here

before i could play at 4K res without stuttering.
now it stutters like crazy
it looks like most 4K videos only plays at 10fps.

also GPU usage is quite big
50% on a GTX 780

any solution to speed up VP9 in Chrome?

I don't see how you could speed up a codec. Instead:

Fake your user agent so that youtube is duped to to think that you're running XP. (I have: "Mozilla/5.0 (Windows NT 5.1) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/40.0.2188.2 Safari/537.36") Then you'll be served AVC streams only.

De-M-oN

29th January 2015, 19:05

Are you sure that you watch a video as VP9? - because still lots of videos dont get VP9 encodes. Like before.

NikosD

29th January 2015, 20:03

Most of the clips are in VP9 now, if not all, and with the first priority.

Of course if you have IE11 or any other incompatible browser, you' ll never see that.

De-M-oN

29th January 2015, 20:08

Most of the clips are in VP9 now, if not all, and with the first priority.

Not really. New uploaded video on my channel and still h.264. You can rightclick in the video and select stats for nerds and you see that it says avc.

Here my video:

https://www.youtube.com/watch?v=ZZSNu0oq-TU&hd=1

No VP9 here.

Nintendo Maniac 64

2nd February 2015, 00:08

Here my video:

https://www.youtube.com/watch?v=ZZSNu0oq-TU&hd=1

No VP9 here.

It says VP9 for me...

De-M-oN

2nd February 2015, 00:12

rofl now it does for me too, wtf xD

But still no 1440p / 4k @ 60fps support :(

Nintendo Maniac 64

2nd February 2015, 06:31

I have: "Mozilla/5.0 (Windows NT 5.1) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/40.0.2188.2 Safari/537.36"

You may want to make your chrome version and webkit versions be really really high numbers so that, in the future, you don't have to be worried about being served and older verison of the page or something.

For example, any useragent overrides I do in Pale Moon I say that my browser is Firefox 99 so that I don't have to worry about websites doing useragent sniffing and serving me older versions or nagging me to upgrade just because they don't know what a "Pale Moon" is (and I don't use the built-in compatibility mode because I specifically want to know if/when a site commits the sin of useragent sniffing).

----------------------------------------------------------------

rofl now it does for me too, wtf xD
I just realized, the video was uploaded on the 29th. As I've stated already, it typically takes about 24 hours for VP9/WebM encodes to appear.

But still no 1440p / 4k @ 60fps support :(
Probably a combination of bandwidth and playback performance, with more importance on the playback performance considering that 4k 60fps would be a great showcase for Google Fiber (seeing how YouTube is owned by Google and all).

EDIT: Yeah, I think it's due to performance; I get 80-100% CPU utilization in Chrome with frames dropping on a 1440p 60fps VP9 video (I took a 1440p 30fps WebM YouTube video and used mkvtoolnix to set the VP9 video stream to 60fps). Now consider that I'm running a Pentium G3258 @ 4GHz.

However, it was perfectly smooth in MPC-HC 64bit with around 60-80% CPU utilization, so maybe with more optimizations to libvpx we could maybe see 1440p VP9 60fps support.

EDIT 2: Using the same mkvtoolnix method to set a 4k 30fps VP9 video to 60fps, I just get constant 100% CPU utilization in MPC-HC 64bit + LAVfilters trunk revision 0.67, so 4k 60fps VP9 is definitely out of the question in the near future. (FYI, this was using "EVR custom" + "D3D Fullscreen" + 'Video Frame' set to "Normal Size")

Yes Google/YouTube could always make an exception and use h.264 for 60fps in resolutions higher than 1080p, but I wouldn't hold my breath...

De-M-oN

2nd February 2015, 16:00

Youtube wants to implement 1440 and 4k at 60fps.

Here is already a test video.

https://www.youtube.com/watch?v=bfUeocvzApg

-> http://www.reddit.com/r/youtube/comments/2maxy6/4k_and_60fps_videos/

The thing is: I'm waiting ... and waiting ... and waiting xDDD --- the general problem with youtube :D

The thing is not just the smoother video why I want it - I want it, because youtube offers more frame quality at 60fps and they offer in general more quality at higher level than only 1080. Their 1080 quality doesnt satisfy me at all - but at least it has better quality at 60fps.

Bitrate increase is 50% at the 60fps variants. I hope it will be the same for 1440 and 4k. Then these will look awesome :) (oh and to get the best out of their 4k quality - use 3200x1800 for 16:9 or 2800x1750 for 16:10. And for 1440p - already 1920x1200 would be enough for 16:10 or 2080x1170 for 16:9. And 1920x1200 is only slightly bigger than 1080p, but the 1440p level has at 30fps a bitrate of around 10000 kbit, while 1080p 30fps has only 4000 kbit. And that of course is a big improvement.
60fps 1080p is at around 6000 kbit though - means an increase of 50%. If 1440p gets then +50% as well - oh well that would be nice :D Imagine 15000 kbit for 1920x1200^^

And JDownloader cant access the 60fps 1440 and 4k of the test video - so I cant check.

sneaker_ger

2nd February 2015, 16:17

Use youtube-dl (http://rg3.github.io/youtube-dl/), it will show all streams (and download them if you want to).
3840x2160 60 fps VP9: 1010 MB
2560x1440 60 fps VP9: 439 MB

De-M-oN

2nd February 2015, 16:28

hmm would have to read how it works before it seems - but looking at your filesizes - its not 50% higher on these like it was at 1080 and 720.

hmpf. But at least a bit higher.

mzso

2nd February 2015, 17:30

Ugh. Why waste resources on nonsense like this. They should double the bitrate for up to fullHD stuff first because they look like crap.

De-M-oN

2nd February 2015, 17:32

Yes but they dont :(

zerowalker

2nd February 2015, 17:44

Oh didn't know bitrate increased in 60fps, thought it was pretty much the same and they just kinda forced it in.
That's good news:D

De-M-oN: 1920x1200, the best aspect ratio;)!

mzso

2nd February 2015, 17:52

Yes but they dont :(

Gah. Just checked: Dailymotion has 2.4 times the bitrate for fullHD...

De-M-oN

2nd February 2015, 18:00

NikosD

2nd February 2015, 20:09

Here is already a test video.

https://www.youtube.com/watch?v=bfUeocvzApg

A very demanding clip even for my Core i7-4790 and Chrome x64 v.41

By selecting 2160p VP9 60fps, I have a lot of dropped frames due to slow decoding in window and full screen mode.

Chrome needs to update the VP9 decoding engine.

Use youtube-dl (http://rg3.github.io/youtube-dl/), it will show all streams (and download them if you want to).
3840x2160 60 fps VP9: 1010 MB
2560x1440 60 fps VP9: 439 MB

Nice app I downloaded the 2160p VP9 60fps clip, but it's video only.
I didn't find a choice of full video+audio stream.
I don't want to download audio and video separately and merge them afterwards.

Using MPC-HC x64 latest nightly (LAV internal), I decoded the clip in 4K VP9 60fps easily with <50% CPU utilization.

De-M-oN

2nd February 2015, 20:34

I don't want to download audio and video separately and merge them afterwards.

You have to. They arent muxed at youtube either (DASH Streaming).

NikosD

2nd February 2015, 20:46

I meant using IDM or other similar apps to download clips from Youtube, you get the muxed video+audio file.

De-M-oN

2nd February 2015, 20:55

I normally prefer jdownloader 2, but it cant find 1440p60, and 4k60 so far.

Nintendo Maniac 64

3rd February 2015, 00:12

Oh, so 1440p and 2160p 60fps VP9 do exist? That might explain what was stated on the CompleteYouTubeSaver MantisHub page on the 31st about "new 60fps formats"...

And guys, about VP9 and audio, MPC-HC has the ability to play MKV and WebM video files with external audio files as long as they have the same filename (other than extension). Since the VP9 and Vorbis DASH streams (no idea about Opus) are already natively in matroska/webm, you don't need to repackage them or anything other than using different file extension (for example, WebM audio is technically .MKA since even foobar2000 can play taht back, and foobar is very picky about file extensions matching their format).

mzso

3rd February 2015, 13:18

Dailymotion uses 6000 kbit for 1080p. Youtube 4000 kbit (at 30fps - but only if the video complexity is enough. A tutorial video for example will never get 4000. My NFS 1 videos often only get half of the bitrate)
Dailymotion though uses always 6000 kbit. And they even encode 2pass ! with x264 preset slow !!

Of course that has by far better quality than youtubes rather very fast encoding settings and 1 passed bitrate limitation and even combined with a high CRF or with whatever method they imply the complexity within the max bitrate.

Well I mostly get 2500kbps There's like 1-2 that's ~3000kbps...
.MKA since even foobar2000 can play taht back, and foobar is very picky about file extensions matching their format).
I always hated that. What is this the eighties?
I meant using IDM or other similar apps to download clips from Youtube, you get the muxed video+audio file.
Well, I guess you have to wait for other programs to be updated with these formats.

mzso

3rd February 2015, 15:41

So 60fps 2160p and 1440p only exists in VP9?
I can't even play vp9 at fullHD 60fps...

dapperdan

3rd February 2015, 17:20

Maybe they wanted test data for upcoming hardware decoders? A chromecast v2 or whatever. It would make sense for them to synce between hardware and the service most likely to use it.

Nintendo Maniac 64

4th February 2015, 00:19

I can't even play vp9 at fullHD 60fps...

Use either the latest trunck builds of LAVfilters or use the latest 64bit MPC-HC. I know with the latter I can do 1080p 60fps VP9 on even a 2.4GHz Core 2 Duo without any dropped frames (but with like 95% CPU utilization :p)

Note however that if you have a CPU without SSSE3 (not SSE3), you may have performance issues (I have not yet tested the latest trunk 32bit LAVfilters on a CPU without SSSE3).

mzso

4th February 2015, 12:02

Use either the latest trunck builds of LAVfilters or use the latest 64bit MPC-HC. I know with the latter I can do 1080p 60fps VP9 on even a 2.4GHz Core 2 Duo without any dropped frames (but with like 95% CPU utilization :p)

Note however that if you have a CPU without SSSE3 (not SSE3), you may have performance issues (I have not yet tested the latest trunk 32bit LAVfilters on a CPU without SSSE3).

Huh. It's a lot better with mpc-hc: 50-60% CPU usage. Though it's hard to tell framedrops without madvr. (Ugh 32/64 bit versions share settings... and falls back to a crappy render from madvr.)
Potplayer 32 with LAV from here (http://files.1f0.de/lavf/nightly/) is not near as fast. Can't play it full speed. I get about 90% cpu usage and framedrops.

NikosD

4th February 2015, 12:29

There is no need for PotPlayer to use other filters/decoders than its own.

Using latest PotPlayer x64 with internal decoders for 2160p60 fps VP9 file, I see the same CPU utilisation like MPC-HC x64 with LAV internal.

mzso

4th February 2015, 13:43

There is no need for PotPlayer to use other filters/decoders than its own.

Except for the many crappy issues... Like the internal (quality destroyer) video processing filter getting enabled. Or bugged decoders with broken output.

NikosD

4th February 2015, 15:10

All decoders have issues now and then.

Which video processing filter ?

mzso

4th February 2015, 16:31

All decoders have issues now and then.

Which video processing filter ?

There's only one. Which by default converts everything to NV12. It doesn't matter that you have it disabled. It gets activated anyway if an internal decoder is loaded, most of which you can't disable.
Also some internal decoders produced garbled output or subpar performance, but you can't disable them.

Nintendo Maniac 64

5th February 2015, 00:10

Potplayer 32 with LAV from here (http://files.1f0.de/lavf/nightly/) is not near as fast. Can't play it full speed. I get about 90% cpu usage and framedrops.

Hmm...may I ask what your CPU is?

mzso

5th February 2015, 01:55

Hmm...may I ask what your CPU is?

e6750. I just assumed that on 32 bit it's just slower.

Nintendo Maniac 64

5th February 2015, 04:04

e6750. I just assumed that on 32 bit it's just slower.

With the trunk LAVfilter builds I mentioned it shouldn't be noticably slower on 32bit...

You could try doing the thing I described in the LAVfilters thread:
Fun fact: the DLLs and stuff in that installer seem to be perfectly compatible with MPC-HC.

In other words, if you run the installer in something like sandboxie, you can copy the contents from the x86 and/or x64 folder into the LAVFilters and/or LAVFilters64 folders in the MPC-HC directory and everything will "just work".

mzso

5th February 2015, 11:42

Well, 32 bit MPC-HC is a fair bit slower too, though not as much.

Nintendo Maniac 64

6th February 2015, 00:23

Well, 32 bit MPC-HC is a fair bit slower too, though not as much.

Stock 32bit MPC-HC is indeed slower, which is why I mentioned the trunck LAVfilters since they include new optimizations for 32bit VP9 decoding.

mzso

6th February 2015, 00:52

Stock 32bit MPC-HC is indeed slower, which is why I mentioned the trunck LAVfilters since they include new optimizations for 32bit VP9 decoding.

Well it's the slowest of the three for me.

Nintendo Maniac 64

6th February 2015, 01:16

Well it's the slowest of the three for me.

*sigh* I don't think you're quite understanding what I'm saying. Alright then, I'll practically do the hard part for you then.

Go into your 32bit MPC-HC folder. Rename the "LAVFilters" folder to "LAVFilters_BAK" or similar.

Download this:
EDIT: I'm a derp, see this post (http://forum.doom9.org/showthread.php?p=1708478#post1708478).

Extract it.

Move the newly-extracted "LAVFilters" folder into your 32bit MPC-HC folder.

Your 32bit VP9 performance should now be on par with your 64bit VP9 performance.

mzso

6th February 2015, 02:39

*sigh* I don't think you're quite understanding what I'm saying. Alright then, I'll practically do the hard part for you then.

Go into your 32bit MPC-HC folder. Rename the "LAVFilters" folder to "LAVFilters_BAK" or similar.

Download this:
http://www.mediafire.com/download/5ub6p014s12mfu1/LAVFilters_0.63.0-74_32bit.zip

Extract it.

Move the newly-extracted "LAVFilters" folder into your 32bit MPC-HC folder.

Your 32bit VP9 performance should now be on par with your 64bit VP9 performance.

I'm not the one who doesn't understand. LAV 32bit was the slowest (as tested in potplayer) so I saw no point in copying it anywhere.
Anyway I can just set the external LAV in MPC-HC. Actually it's faster in MPC-HC (somethings sucks with potp) around on par with the internal.
So there. The 64 bit internal LAV/MPC-HC is just faster on my system.

zerowalker

6th February 2015, 02:42

mzso

6th February 2015, 02:47

Confused that the debate is about.

You are saying VP9 is slow in 32bit and fast in 64bit (which is normal right?).

Nintendo is saying you can use the new LAV instead of Stock that has improved speed (i think?).

But you are saying that Stock is faster than the new one?

Circle of doom;O

No I said that the internal of the latest MPC-HC and the latest LAV nightly is around the same. (With potplayer it's slower for some reason)
64 bit is faster.

zerowalker

6th February 2015, 03:10

Ah ok.

So, the misunderstanding is Nintendo trying ot make you use something that doesn't help:P?

Well never used potplayer myself, but can't you just use MPC-HC 64bit and be happy?
Though then again i use Zoom Player which is only 32bit so if you prefer the other player i feel you.

Nintendo Maniac 64

6th February 2015, 04:04

mzso

6th February 2015, 12:49

Wait a minute, what version of MPC-HC are you using?

Turns out the newest update (v1.7.8) includes newer LAVFilters with said 32bit VP9 optimizations. The whole time I've been running on the assumption that v1.7.7 was still the newest version.

I got it from here: http://nightly.mpc-hc.org/
So I guess the freshest.

Ah ok.

So, the misunderstanding is Nintendo trying ot make you use something that doesn't help:P?

Well never used potplayer myself, but can't you just use MPC-HC 64bit and be happy?
Though then again i use Zoom Player which is only 32bit so if you prefer the other player i feel you.

Oh I'm happy I just don't watch VP9 videos. :) I just mentioned that I can't even play FullHD at 60fps.
Youtube will provide AVC for the foreseeable future, which is for the most part of similar or better quality (because of the higher bitrate) with few exceptions. Plus no-one else uses it.
If for some reason I need to watch a fullhd vp9 video I'll just watch in mpc-hc x64

Anyway it's unusual to for PP to perform worse than mpc-hc with the same filters. It might be related to the fact that the only way PP will use a system codec if you remove ffcodec.dll from it's folder. And that's not quite standard usage.

zerowalker

8th February 2015, 18:47

Well you aren't alone.
I can watch Full HD and all, but it still uses A Lot of resources.
The difference between VP9 and H264 similar or better quality isn't even comparable.
The speed difference is far beyond the quality difference and still VP9 is pretty much on par to x264 more or less (compared to x265 which has a reason for being slow).

I have no idea why it's so slow, even VP8 is slow and it looks worse.

I still like it though, new approaches and competitions brings forth development;)

Nintendo Maniac 64

9th February 2015, 02:08

The speed difference is far beyond the quality difference

It shouldn't be that much slower if you're using 64bit MPC-HC and both codecs are being decoded in software without any hardware acceleration.

Using the following video as a test (1080p 60fps):
https://www.youtube.com/watch?v=5I-nQtsJuhE

I get around 30-40% CPU utilization for VP9 and 25-35% CPU utilization for h.264 without hardware acceleration. My CPU is a Pentium G3258 @ 3.8GHz.

zerowalker

9th February 2015, 15:50

Meant Chrome, 64bit is indeed much faster, but still not worth the difference tbh.
But if it brings quality to Youtube in some degree it's kinda worth it in that way;D

xooyoozoo

17th February 2015, 02:14

Someone made a Chrome extension to force YouTube to play H.264/AVC instead of VP9. (https://github.com/erkserkserks/h264ify)

Edit: Direct link to its Chrome Web Store. (https://chrome.google.com/webstore/detail/h264ify/aleakchihdccplidncghkekgioiakgal)

mzso

17th February 2015, 12:44

De-M-oN

17th February 2015, 17:20

Magic Actions can do this as well. And has so much other nice stuff.

http://chromeactions.com/

NikosD

30th March 2015, 09:13

A few 4K60fps VP9 clips from YouTube, which can only be viewed on Google Chrome.

https://www.youtube.com/playlist?list=PLyqf6gJt7KuHCoYaGNWPx8vkHlClmbO9u

littleD

30th March 2015, 10:01

Cool. At least I don't have to rely on user agent faking.Thats a great option too: Viewtube script (http://isebaro.com/viewtube/?ln=en) You have great control over what you see on youtube. Only issue I cannot see by this script UHD videos. If they are dash only, then current version of Firefox might have problem with it.

Nintendo Maniac 64

31st March 2015, 04:07

A few 4K60fps VP9 clips from YouTube, which can only be viewed on Google Chrome.

https://www.youtube.com/playlist?list=PLyqf6gJt7KuHCoYaGNWPx8vkHlClmbO9u

Huh, you're right - it's not even available in Firefox Nightly 39. The only other way is to use a YouTube downloader of sorts.

EDIT:
Only issue I cannot see by this script UHD videos. If they are dash only, then current version of Firefox might have problem with it.

480p and 1080p are DASH-only as well, yet Firefox can still see those. I wonder if Google/YouTube is using a higher VP9 encoding setting or something for the UHD videos than Gecko currently supports?

Easiest way to find out would be to download a video and try to play it back directly in the browser...I'm going to go test that now. My connection isn't the fastest at all (3Mbps down), so it'll take a a bit of time...

EDIT 2: Firefox 36.0.4 was able to play the downloaded 1440p 60fps and 2160p 60fps video just fine, so I don't know what the deal is...

mzso

31st March 2015, 11:27

EDIT 2: Firefox 36.0.4 was able to play the downloaded 1440p 60fps and 2160p 60fps video just fine, so I don't know what the deal is...

Something's definitely messed up for a while now.

https://www.youtube.com/watch?v=hicBgE6XndM

For this video for example only 1080p shows up in the player. After a while you can see it in the statistics that it switches to 1920*1200 (the original resolution), but the player still shows 1080p.
I noticed other videos missing 1920*1200, but some (with a lot of views) had it show up in the player(as 1440p).

dapperdan

31st March 2015, 11:33

I don't think I see 480p and 1080p in my Firefox (Mac OS X, no Flash installed) just 360 and 720.

I thought the status was that Firefox supported enough of MSE for some DASH demos to work, but not yet enough for Youtube to work reliably? Presumably they can play the files, just not switch between them via js in response to bandwidth fluctuations

Google have a semi-public test page for this, in firefox for me it bails after testing for MSE:

http://dash-mse-test.appspot.com/dash-player.html

mzso

31st March 2015, 13:17

littleD

31st March 2015, 14:02

Looks like youtube gives different sources with Uhd material. Maybe not all links are problematic. Here it works with current firefox with no registry changes https://www.youtube.com/watch?v=Tz6RxJtS8Kc or https://www.youtube.com/watch?v=92aGoXacpnE

De-M-oN

31st March 2015, 15:51

For this video for example only 1080p shows up in the player. After a while you can see it in the statistics that it switches to 1920*1200 (the original resolution), but the player still shows 1080p.
I noticed other videos missing 1920*1200, but some (with a lot of views) had it show up in the player(as 1440p).

It is just bugged for their html5 player.

In Flash Mode you'll see all quality levels.

For HTML5 1080p is gone and the 1440p quality (in this case 1920x1200) is running.

Same for 3200x1800. No access to 1440p, instead on 1440p runs the 4k file.

But thats only with HTML5 the case at the moment.

mzso

31st March 2015, 16:51

It is just bugged for their html5 player.

In Flash Mode you'll see all quality levels.

For HTML5 1080p is gone and the 1440p quality (in this case 1920x1200) is running.

Same for 3200x1800. No access to 1440p, instead on 1440p runs the 4k file.

But thats only with HTML5 the case at the moment.

That "their" is probably youtube though, whit chrome it's also messed up if memory serves.

De-M-oN

31st March 2015, 16:54

Yes I meant youtube.

dapperdan

7th April 2015, 12:00

Some PR fluff from Youtube about their use of VP9:

http://youtube-eng.blogspot.co.uk/2015/04/vp9-faster-better-buffer-free-youtube.html

Only interesting tidbit is the focus on people with half-decent desktop machines but terrible internet speeds getting upgraded to SD resolutions due to VP9.

marauder

7th April 2015, 17:10

Anecdote example from a user living in developing country:

VP9 has made huge difference in my video viewing experience.

Being on a 1.5 mbps DSL I could only watch in 480p if I wanted to watch youtube without buffering.

Buffering / downloading time for 720p h264 is 2-3 times the running time and since I almost exclusively watch long videos (averaging 40 minutes) it is not practical to watch on the fly and I had to resort to queuing several download before going to bed and watch them the next day on Potplayer.

Now I can watch on 720p by letting the video buffer for a couple minutes. It's awesome and though the switch to VP9 had begun a while ago I hadnt noticed until the announcement. Guess Video Downloader extension will be seeing less use.

STaRGaZeR

7th April 2015, 20:34

Now imagine if instead VP9 they used properly encoded H.264. You could even have hardware acceleration and even more quality! :D

Progress...

dapperdan

7th April 2015, 22:53

Nintendo Maniac 64

7th April 2015, 23:30

It's my understanding that, at low bitrates (~1Mbps), VP9 is better than x264 , but the opposite is true for high bitrates.

Also, now I feel terrible that I don't even live in a developing country and I only get 3Mbps DSL. Good ol' U S of A, more like the land of free handouts for mono/duopolies.

Tommy Carrot

7th April 2015, 23:32

I'm guessing that the smiley means you're joking, but I'm interested to know what the general consensus is these days on VP9 vs x264 for this usecase? It seems like everyone got bored of publishing codec comparisons a while ago. So where does it stand these days? Does anyone know?

In my opinion it's the same deal as with x265. It's better than x264 for low bitrates, but x264 is better for high bitrates. Of course it varies with the type of content, but in my experience if you want to encode with x264 over crf 22-24, for similar bitrate you will get better quality with vp9 (and x265), below that x264 seems better.

Nintendo Maniac 64

8th April 2015, 03:36

I just did an encoding test at 2125Kbps (the bitrate used for 720p on YouTube) with some 720p F-Zero GX footage I recorded a year ago. Personally I'd say that with high-speed motion, VP9 looks more natural since the areas of not-enough-bitrate end up being somewhat blurred rather than having visible macro-blocking.

By comparison, with x264 at the same bitrate, the areas of not-enough-bitrate give an annoying shimmery texture aliasing-like-effect similar to a 3D-polygonal game without mipmapping that's only using bilinear texture filtering - and this is even with x264 configured with unrealisticly high settings (read: placebo).

This was the source video in question that I encoded from (was admittedly one of my earlier encodes, so the x264 settings used weren't the greatest):
http://www.mediafire.com/download/s525ycmc05nbohs/C_-_1280x720_@_30fps.mp4

xooyoozoo

8th April 2015, 04:09

Turn off psy-rd.

I want to say that the usage of psy is barely relevant for streaming companies, but Vimeo seems to do fine with it, so who knows.

vivan

8th April 2015, 10:09

Because vimeo uses sane settings and bitrates :P

This was the source video in question that I encoded from (was admittedly one of my earlier encodes, so the x264 settings used weren't the greatest):
http://www.mediafire.com/download/s525ycmc05nbohs/C_-_1280x720_@_30fps.mp4Source already seriously lacks bitrate. At 2 mbps all encodes look like complete and utter shit. It starts to become reasonable only at 360p (which is --crf 22).

source http://firepic.org/images/2015-04/08/6xknz8zhy63r.png
vp9 http://firepic.org/images/2015-04/08/b4pqvqfyor0s.png
720p http://firepic.org/images/2015-04/08/4y1i61amnwlc.png
360p http://firepic.org/images/2015-04/08/9dsn92lrp20h.png

source http://firepic.org/images/2015-04/08/s174ium5bv05.png
vp9 http://firepic.org/images/2015-04/08/j7yja25sajzc.png
720p http://firepic.org/images/2015-04/08/tv0avfkcfqea.png
360p http://firepic.org/images/2015-04/08/smru6rf2bf3u.png

And this is the youtube problem - every resolution is a bitstarved mess that looks worse than 2x lower res at the same bitrate. What they want is numbers (4K, 60 fps sounds so cool), not quality.

mzso

8th April 2015, 11:35

dapperdan

8th April 2015, 18:34

When people say the next-gen codecs win at low bitrate, they mean low bit-per-pixel, right? So low bitrate for 4K is going to be a higher number than low bitrate for 720p?

And does that mean that as x264 starts to win as the bitrate increases you should perhaps be comparing it against a higher resolution encode using the same bitrate?

The scalability of web video adds another dimension to codec comparisons, and it kind of makes my head hurt to think about it. Is this covered by the objective metric graphs that you see, or do they all assume a fixed resolution? I think I've read that some of the objective metrics fall apart entirely if you resample in this way. It also sounds like exactly this is what is being complained about above ("every resolution is a bitstarved mess that looks worse than 2x lower res at the same bitrate") but is that actually true? When does it make sense to switch resolution up or down from a quality perspective? Is there a standard rule of thumb?

STaRGaZeR

8th April 2015, 19:02

I'm guessing that the smiley means you're joking, but I'm interested to know what the general consensus is these days on VP9 vs x264 for this usecase? It seems like everyone got bored of publishing codec comparisons a while ago. So where does it stand these days? Does anyone know?

Not joking at all. The consensus is that at sane bitrates x264 encodes win hands down plus the hardware acceleration possibility (or should I say reality, because it's used everywhere), and that the encodes of almost all streaming sites suck balls regardless of the format used.

dapperdan

8th April 2015, 20:46

Is there a plausible reason why encodes of streaming sites suck? You'd think it was in their business interests to be good at this kind of thing, but it's fairly unanimous from people on sites like this that they're doing it wrong. Is it because, as with mp3s, the average consumer doesn't really care and so there's no profit motive in pushing quality?

Even then you'd think they'd use the best encoding technology they can, and just keep lowering the bitrate till people disliked it enough that it affected ad revenue. Or has that already happened?

captainadamo

8th April 2015, 21:07

Most people won't care. So with the sheer amount of content they are dealing with, spending the extra CPU cycles for extra quality is probably seen as a waste.

STaRGaZeR

8th April 2015, 21:49

People don't care indeed. You put a 1080p logo below the video and it doesn't matter if it looks like shit, people say ohhhh 1080p! Amazing! Just like with MP3 but much worse IMO. The content itself matters much more than the quality for the majority of the target audience.

About wasting CPU cycles on more quality... forget about that, you have to wonder why they waste CPU cycles on LESS quality with VPx :D

Nintendo Maniac 64

8th April 2015, 23:14

Come on now, there are times when the resolution is beneficial even at low bitrates. Considering the proliferation of let's plays, in-game text greatly benefits from increased resolution even at said low bitrates.

And regarding the lack of bitrate with my sample recording, I did warn you that it wasn't a great encode.

De-M-oN

8th April 2015, 23:25

Wasting traffic and cpu is youtube's hobby ;D

1920x1200 gets their 1440p level though. It has 10000 kbit instead of 4000 kbit at 1080p and that of course looks a lot better.

The 60fps encodes are better bitrate. If you dont use the full 60fps you can even more improve the quality.

For example 45fps is enough to get the HFR bitrate encodes and saves bitrate / frame.

The 60fps are a good addition - this way you can at least control a little bit the smoothness and quality ratio. So if you are way too unhappy with the quality - you even could encode your own encode to 25 fps, set x264 to keyint infinite and then bloat it up to 45 fps (maybe even less is necessary to get the HFR encode. Dont know the minimum at the moment)
The filesize will be exactly the same as a 25 fps encode then.
And with that the complexity - and it will improve quality a lot then at youtube.
Funfact: 1080p gets at HFR Encode about 5000 kbit for video. That was the bitrate you got at around 2011 on the normal 30fps encodes. With 720p the same story.

There is one video portal which indeed offers awesome quality:

Dailymotion

They use x264 @ preset slow a bitrate of 6000 and 2passed. That looks very good in most cases.

Youtube @ 1080p:
http://abload.de/img/desktop_2015_02_17_171qqkl.png

Same Video at Dailymotion on 1080p:
http://abload.de/img/lpt-doom2-oblige0006-3tq4e.png

The youtube result on this video was just a disappointment.
I could get at least about ~90% of the quality of dailymotion by using the 1920x1200 @ 25 fps bloated up to HFR method.

But if you do something like this I highly recommend that you write in videotitle the real fps. It would otherwise fool the viewers and would help the generation of people who say HFR cant be seen.

Youtube uses 4k and such, but using only there a better bitrate. same with HFR and so on. For 25/30fps I recommend either 4k @ 3200x1800 for best image quality or the bloating up to HFR method.
I hope that they finally activate HFR for 1440p. 1440p looks quite ok with 1920x1200 and will look better with the HFR bitrate. And imagine the Bloat to HFR method on 1440p - that would be great. But I will use the best balance out of it - because the smoother fpsrate is liked by people too.

If they would use at least the same bitrate at VP9 like with h.264 - but no - you see what is the goal - higher numbers like 4k and such but with the main goal reduce traffic.. They dont care about quality at all.

And this is the youtube problem - every resolution is a bitstarved mess that looks worse than 2x lower res at the same bitrate. What they want is numbers (4K, 60 fps sounds so cool), not quality.

So yeah - sad but true...

They want numbers and low traffic. But they produce traffic by using shit encodes at 1080p because people then do upscale things to get their better quality levels - so I dont get their strategy. People tend to bloat up to 3200x1800 to get their 4k encode which is at least much better quality than their 1080 and by doing this user and portal has to do more cpu and traffic work - strange strategy by google.

foxyshadis

10th April 2015, 00:37

De-M-oN

10th April 2015, 01:09

unless one just happens to fall on a keyint boundary

Thats why I said : Use keyint infinite. This way this wont happen then.

I would do both though. Youtube will in future allow 1440p60 and 4k60 - its just not enabled for public so far. But there exist videos of it already. And of course at HFR bitrate they look even better then. Youtube will add the missing 1440p60 and 4k60 to videos already uploaded as soon as it is activated for all.
Same happened to 720p60 and 1080p60. A video of 2011 (!) was uploaded at 59,94 fps by the uploader and got now a 720p HFR encode. That was really impressive to me to see how far back they apparently already did HFR encodes.

zerowalker

10th April 2015, 01:16

I think it depends on popularity, like with VP9.
My old 60fps are still 30fps, only those that happened to become VP9 turned to 60fps.

foxyshadis

10th April 2015, 09:46

Thats why I said : Use keyint infinite. This way this wont happen then.

You can't control where YouTube's re-encodes put their keyframes, though. For your original upload you'd encode it much smarter than the tubes, for sure.

De-M-oN

10th April 2015, 11:30

Believe me - The quality will be better this way by far. I tested it several times already.

If you use 25fps bloated up to HFR their 1080p even looks nearly like their 1440p quality @ 1920x1200

zerowalker

10th April 2015, 11:35

So HFR is controled by keyframes or bitrate, confused?
I got 60FPS in 2D games at about 2-4k bitrate, but that's lossless x264 i444.

mzso

10th April 2015, 15:05

Believe me - The quality will be better this way by far. I tested it several times already.

If you use 25fps bloated up to HFR their 1080p even looks nearly like their 1440p quality @ 1920x1200

So what is the minimum for 1440p? Do they measure vertical resolution or pixel numbers?
This video (https://www.youtube.com/watch?v=e9qvH69prjA) for example has two 1600*1200 streams which show up in downloader tools under different names. The 670MB stream as 1440p (or Ultra HD in the other tool), The 400MB stream as 2160p (or Original). The namings are kind of contradictory. :) And it's really weird that two streams with a massively different bitrate are created for the same resolution

De-M-oN

10th April 2015, 15:31

1920x1200 is enough for 1440p
1600x1200 as well.
2080x1170 would be 16:9 and is enough.

They measure both, but you reach the height earlier.

[info] Available formats for e9qvH69prjA:
format code extension resolution note
171 webm audio only DASH audio 115k , audio@128k (44100Hz), 12.78MiB
140 m4a audio only DASH audio 130k , m4a_dash container, aac @128k (44100Hz), 15.93MiB
141 m4a audio only DASH audio 256k , m4a_dash container, aac @256k (44100Hz), 31.64MiB
278 webm 192x144 DASH video 82k , webm container, VP9, 13fps, video only, 8.66MiB
160 mp4 192x144 DASH video 116k , 13fps, video only, 13.68MiB
242 webm 320x240 DASH video 158k , 25fps, video only, 12.22MiB
133 mp4 320x240 DASH video 250k , 25fps, video only, 30.53MiB
243 webm 480x360 DASH video 309k , 25fps, video only, 23.67MiB
134 mp4 480x360 DASH video 322k , 25fps, video only, 30.34MiB
244 webm 640x480 DASH video 550k , 25fps, video only, 43.28MiB
135 mp4 640x480 DASH video 713k , 25fps, video only, 65.05MiB
302 webm 960x720 DASH video 1523k , VP9, 50fps, video only, 131.42MiB
298 mp4 960x720 DASH video 2035k , h264, 50fps, video only, 232.24MiB
303 webm 1440x1080 DASH video 2725k , VP9, 50fps, video only, 230.80MiB
299 mp4 1440x1080 DASH video 3568k , h264, 50fps, video only, 390.87MiB
138 mp4 1600x1200 DASH video 3751k , 25fps, video only, 364.80MiB
271 webm 1600x1200 DASH video 4052k , 25fps, video only, 335.80MiB
264 mp4 1600x1200 DASH video 6112k , 25fps, video only, 642.67MiB
17 3gp 176x144
36 3gp 320x240
5 flv 400x240
43 webm 640x360
18 mp4 640x360
22 mp4 1280x720

This video is not complex enough, it doesnt fill their bitrate maximum.

It feels like they use a combination of a very high crf and a vbv-max.

But its indeed strange that there are for this video two 1600x1200 encodes of h.264.

264 seemed to be used at "original". I checked my older doom 3 2048x1152 videos and at 2048x1152 it was the quality level "original" at that time and the ID 264 seems to be this encode.

mzso

10th April 2015, 15:43

What sort of info tool is this output from?

vivan

10th April 2015, 15:44

youtube-dl (https://github.com/rg3/youtube-dl/) -F URL

mzso

10th April 2015, 16:05

youtube-dl (https://github.com/rg3/youtube-dl/) -F URL

Thanks.

zerowalker

10th April 2015, 16:59

Oh, i was sure they only cared about height for resolution.
so ???x720 = 720p, ???x1080 = 1080p, and >= 1200 = 1440p etc.

I have always used that and it has always worked (but then again have never tried any extreme resolution, but a weird one is 800x720).

Nintendo Maniac 64

10th April 2015, 23:15

Oh, i was sure they only cared about height for resolution.

They used to, but this changed sometime before the last 6 months.

Example videos that are in fact 1920x804 (there's no letterboxing in the actual videos themselves) but still have 1080p quality available:
https://www.youtube.com/watch?v=unaP8sYGO2E
https://www.youtube.com/watch?v=9QFtXq5VknY

zerowalker

11th April 2015, 11:43

Interesting.
But wait, is this the why 1920x1200 is now "1080p"?
Do that mean i get 1080p bitrate instead of the old 1440p?

mzso

11th April 2015, 14:46

Interesting.
But wait, is this the why 1920x1200 is now "1080p"?
Do that mean i get 1080p bitrate instead of the old 1440p?

That's just broken on youtube's side. It actually changes to 1920*1200 covertly after a while... You can see it in the statistics.

zerowalker

11th April 2015, 15:34

Yeah i know, but i mean does that mean that 1920x1200 uses the "1080p bitrate" instead of 1440p which it was used before (i think?).
Or is it simply a title mistake and nothing more?

TheSkiller

11th April 2015, 17:07

My old 60fps are still 30fpsI had the same problem. You can fix it by triggering a re-encode. Go to your video and apply some useless effects. I usually increase the brightness to a silly value. Wait until the processing finished then simply revert to original and it'll process the video again. There you go.

zerowalker

11th April 2015, 17:09

I had the same problem. You can fix it by triggering a re-encode. Go to your video and apply some useless effects. I usually increase the brightness to a silly value. Wait until the processing finished then simply revert to original and it'll process the video again. There you go.

Interesting, will this use the original file then (not some kind of re-re-encode thing)?

And i guess all stats will remain the same?

This wouldn't happen to trigger VP9 as well;P?

TheSkiller

11th April 2015, 17:15

zerowalker

11th April 2015, 17:17

Yes, the original upload will be re-encoded and stats aren't affected.

I actually even noticed an increase in quality because in past years YouTube used to apply unnecessary deinterlacing to all of my x264 encoded uploads (even though correctly flagged as progressive!) which they now don't do anymore.

Didn't check that.

Interesting, when is this "Past"?

Will do a test on a 60fps file, this will be fun;P

EDIT: for some reason i don't have "Enhancements" on that particular file, weird..

EDIT 2: Noticed that my other 60fps are actually 60fps now (and those have the "Enhancement" option), but my oldest videos doesn't.

TheSkiller

11th April 2015, 17:34

Interesting, when is this "Past"?Well, I haven't uploaded anything in the past 11 months (edit: except some recent tests), but they still did it in May 2014.
It's quite obvious if you download such video at it's highest res and look at it frame by frame – whenever there's fine vertical detail appearing (say, a sharp subtitle) it looks exactly like Yadif was applied to it. This never happened if I used Xvid and AVI, only with x264 and MP4/MKV.

zerowalker

11th April 2015, 17:50

Hmm, it's weird that i myself haven't noticed anything. I usually see those kind of artifacts as i find them very annoying, disappointed at myself;P
Will go through some old videos.

mzso

11th April 2015, 19:19

Yeah i know, but i mean does that mean that 1920x1200 uses the "1080p bitrate" instead of 1440p which it was used before (i think?).
Or is it simply a title mistake and nothing more?

I highly doubt it. The download tools shows an 1920*1200 stream that's double the size of the 1080p one:

http://abload.de/img/yt-streams9zkkh.png

zerowalker

11th April 2015, 20:12

Nice, as i got that resolution and the difference from 1080p in minimal, that huge bump in bitrate is spectacular;D

De-M-oN

11th April 2015, 21:44

But wait, is this the why 1920x1200 is now "1080p"?

It isnt. Use Flash Player and you'll have it like before.
For HTML5 it is at the moment that the 1080p is inaccessable and in place of 1080 the 1440p file is playing.

Same for a 3200x1800 video:
At HTML5 4k is missing and replaces 1440p.

Yes, the original upload will be re-encoded

Source? Test? I highly doubt that youtube is this intelligent and does it like this xD

Nintendo Maniac 64

11th April 2015, 23:12

Source? Test? I highly doubt that youtube is this intelligent and does it like this xD

I have some 4-5 year old 1440p and 2400p videos on my YouTube channel that I can test with...

zerowalker

11th April 2015, 23:24

Oh, this sucks if it's not the original that's re-encoded:(
Well i downloaded the last Youtube re-encode so i will be able to compare.
But i actually think it uses the original, as Youtube stores the source it use that for all encodes,
i doubt it actually re-encoded an encoded file when it has the source laying around.

De-M-oN

11th April 2015, 23:35

If they dont delete the source though - why arent we offered to download the source again? :/
At Dailymotion you can download your source at any time - and also their encodes. And you see for every encode the progress percent. At youtube only the progress of the first quality..

zerowalker

11th April 2015, 23:41

We are, we can download the source again, though i can't (or rather it's so huge amount of data that the archive fail).
You have to use Google Data Tools, it's a hassle, i have no idea why they don't allow you to download the file directly (i think illegal material are not available though, the source is probably stored but i think you aren't able to get it).

zerowalker

12th April 2015, 01:25

Okay it's now done. I the difference is close to none, i would say that the new one appears to be a bit better in the cases where it's noticeable though.
But then again there is a bitrate difference, 5931kbps on the first version, 6877kbps on the second.
So pretty sure that explains it, 1mbps more is nothing i say no to;P

No Interlacing bug on first or after btw, the old encoding was from "2013-09-19", new one "2015-04-11". (This according to the downloaded files via MediaInfo, the upload date is from 2011 on the source).

EDIT:

Wait the interlaced bug may have been there, cause i think there was some line bugs in flashplayer that i can't see on any version that i downloaded.
I thought it was just bad text but pretty sure it as this "-" at the top of the letters.
So if the bug exists, does it only affect flashplayer? (using LAV Video and it seems to be progressive for me).

Video
ID : 1
Format : AVC
Format/Info : Advanced Video Codec
Format profile : High@L5
Format settings, CABAC : Yes
Format settings, ReFrames : 2 frames
Codec ID : avc1
Codec ID/Info : Advanced Video Coding
Duration : 35mn 50s
Bit rate : 5 928 Kbps
Width : 1 920 pixels
Height : 1 200 pixels
Display aspect ratio : 16:10
Frame rate mode : Variable
Frame rate : 30.000 fps
Minimum frame rate : 30.000 fps
Maximum frame rate : 3 000.000 fps
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 8 bits
Scan type : Progressive
Bits/(Pixel*Frame) : 0.086
Stream size : 1.48 GiB (100%)
Encoded date : UTC 2013-09-19 15:52:53
Tagged date : UTC 2013-09-19 15:52:53

Video
ID : 1
Format : AVC
Format/Info : Advanced Video Codec
Format profile : High@L5
Format settings, CABAC : Yes
Format settings, ReFrames : 3 frames
Codec ID : avc1
Codec ID/Info : Advanced Video Coding
Duration : 35mn 50s
Bit rate : 6 874 Kbps
Width : 1 920 pixels
Height : 1 200 pixels
Display aspect ratio : 16:10
Frame rate mode : Constant
Frame rate : 30.000 fps
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 8 bits
Scan type : Progressive
Bits/(Pixel*Frame) : 0.099
Stream size : 1.72 GiB (100%)
Encoded date : UTC 2015-04-11 23:19:05
Tagged date : UTC 2015-04-11 23:19:05

Nintendo Maniac 64

12th April 2015, 23:08

I'm in the process of determining the minimum resolution necessary for gaining access to a quality higher than 1080p. Currently I've barely done anything, but I can say that 1922x1080 nor 1920x1082 aren't enough (only 1080p is made available).

De-M-oN

13th April 2015, 03:16

https://www.youtube.com/watch?v=j2Zhyw-RHec

He did the test, but I unfortunately cant find the playlists again.

If I remember right you need at least a height of 1154 or sth like that.

___

The 1080p deinterlace bug is solved since years. Wouldnt bother with it further.

TheSkiller

13th April 2015, 14:28

I'm in the process of determining the minimum resolution necessary for gaining access to a quality higher than 1080p.I did some tests yesterday about the same subject. Not sure if it is the minimum but a height of 1152 is enough to trigger the ~10 Mbps "1440p" encode (which indeed looks a lot better than ~4 Mbps standard 1080p).

However, here's the thing, for whatever reason it was still labeled as 1080p, although it is in fact 1152 at 10 Mbps and there was no regular real 1080p instead (!). In case of my 1152 upload the regular 1080p was dropped in favour of the original (1152) but YouTube still calls it 1080p nevertheless.
You can tell by right clicking on the video player and select "Stats for nerds" (or download the 1080p video).

I also noticed >1080p does not seem to work if the uploaded video has a frame rate greater than 30 fps.

Currently I've barely done anything, but I can say that 1922x1080 nor 1920x1082 aren't enough (only 1080p is made available).Check if it's really downsized to 1080p or if it's left at the original size but still called 1080p.

The 1080p deinterlace bug is solved since years.Cannot confirm it has been solved for years. The deinterlacer was changed from duplicating a field to something like Yadif years ago which made the bug much less noticeable. But it certainly went completely gone only within the last 11 months.

De-M-oN

13th April 2015, 14:47

It is only labeled at HTML5 Player wrong. For whatever reason^^
But to me it is rather a bit funny. Because people with no knowledge often assume they dont gain from higher resolution than their monitor does support - they totally forget frame quality.
But this way these people think they clicked 1080p and play in fact the better file ;D

I also noticed >1080p does not seem to work if the uploaded video has a frame rate greater than 30 fps.

So far not for public. But there are testvideos out and it just takes time until it is available for all people (like it was with the 60fps implementation in general as well)
They said they will offer newly uploaded HFR videos the missing 1440 and 4k ones as soon as it is activated for public.

TheSkiller

13th April 2015, 14:53

It is only labeled at HTML5 Player wrong.I don't know, never used HTML5 player. It's labeled as 1080p in Flash too.^^

Have a look at my 1152 test video.
https://youtu.be/oNdomcZm5QI

Does it play at 1152 and ~10 Mbps for you as well if you select 1080p?

De-M-oN

13th April 2015, 15:13

2080x1170 should be labeled as 1440p for flash though.

TheSkiller

13th April 2015, 15:24

So somewhere between 1152 and 1170 the actual 1440p option is added and if you use 1152 it currently replaces 1080p with a 1440p quality encode.

stax76

13th April 2015, 15:26

I'm trying to add VP9 support to StaxRip but not very successful so far. With ffmpeg I get a crash in the second pass, ffmpeg authors said to report it to libvpx authors which said I should update to libvpx 1.4, I'm stuck here because I don't know how to get ffmpeg with libvpx 1.4.

I also tried vpxenc posted by Kurtnoise recently but this don't seem to support avs import, I tried piping with ffmpeg and avs2pipemod, IIRC with ffmpeg it starts encoding but crashes after a while, with avs2pipemod it's not starting.

Kurtnoise

13th April 2015, 16:29

Using my own build and avs2yuv, the pipeline works fine for me...you should post the command line that you used.

btw, this is not the correct thread. This one should be better (http://forum.doom9.org/showthread.php?t=165839&page=33)...

Nintendo Maniac 64

14th April 2015, 04:05

I just uploaded a 1920x1152 test video:
https://www.youtube.com/watch?v=92hwJoQFxT8

If you use a YouTube downloader, you'll see that both the 1080p and the 1440p formats are available, so the HTML5 player only listing one or the other is just a player glitch rather than a bug with YouTube's encoder backend.

EDIT: So far I know that it has to be larger than 1134p, but oddly enough uploading at that resolution resulted in a choppy video (~15fps real framerate) and made 60fps available though the source was 30fps...:
https://www.youtube.com/watch?v=qzBzVU4ckWU

EDIT 2: ...but uploading at 1136p results in only getting 360p? O_o:
https://www.youtube.com/watch?v=bnalwhUZ0w4

EDIT 3: Well it doesn't matter that much since Ultra HD still isn't made available even at 1150p:
https://www.youtube.com/watch?v=3513LjzB9L4

So it looks like 1152p is indeed the lowest vertical resolution where the Ultra HD formats get encoded. Now we just have to test horizontal resolution...

mzso

14th April 2015, 10:11

Maybe you could report these encoding failures to youtube.

TheSkiller

14th April 2015, 13:29

EDIT 2: ...but uploading at 1136p results in only getting 360p? O_o:
https://www.youtube.com/watch?v=bnalwhUZ0w4I get up to 1080p on that video. And selecting 1080p results in playback of 1826x1080 at normal ~4 Mbps bitrate for me.
I think you need to wait a little longer for all the options to become available.

Now we just have to test horizontal resolution...All my tests were 4:3 resolutions (like 1536x1152), and it worked just like it does with 16:9. I don't think horizontal resolution matters as long as you stay within reasonable limits.

mzso

14th April 2015, 13:53

I get up to 1080p on that video. And selecting 1080p results in playback of 1826x1080 at normal ~4 Mbps bitrate for me.
I think you need to wait a little longer for all the options to become available.

All my tests were 4:3 resolutions (like 1536x1152), and it worked just like it does with 16:9. I don't think horizontal resolution matters as long as you stay within reasonable limits.

Did you try things like 1080*1920 or 1152*2048, so 9:16 inverted aspect ratio stuff that might come from cellphones?

De-M-oN

14th April 2015, 19:57

If you use a YouTube downloader, you'll see that both the 1080p and the 1440p formats are available, so the HTML5 player only listing one or the other is just a player glitch rather than a bug with YouTube's encoder backend.

How already said: For HTML5 1080p gets replaced by the 1440p file

However, here's the thing, for whatever reason it was still labeled as 1080p, although it is in fact 1152 at 10 Mbps and there was no regular real 1080p instead (!). In case of my 1152 upload the regular 1080p was dropped in favour of the original (1152) but YouTube still calls it 1080p nevertheless.
You can tell by right clicking on the video player and select "Stats for nerds" (or download the 1080p video).

For 2080x1170 1080p is back for flash player though.

Nintendo Maniac 64

14th April 2015, 23:14

I get up to 1080p on that video. And selecting 1080p results in playback of 1826x1080 at normal ~4 Mbps bitrate for me.
I think you need to wait a little longer for all the options to become available.

All I know is that the 4 tests that I uploaded after that (1144p, 1148p, 1150p, and a second 1152p test) all had 1080p while that one still had 360p.

Motenai Yoda

15th April 2015, 01:17

err now youtube goes on adaptive bitrate... I don't know what's their goal but now I get up to 8Mbps even for some 240p.
btw 720p for vp9 but 1080p for h.264

De-M-oN

15th April 2015, 01:40

I don't know what's their goal but now I get up to 8Mbps even for some 240p.

Show me videolink please. Not that you measured your downloadspeed. Anyway - I'd like to see that.

Nintendo Maniac 64

15th April 2015, 23:51

There haven't been any new VP9 encodes in 2-3 days...I wonder if they're updating the encoder backend.

De-M-oN

16th April 2015, 00:59

Motenai Yoda

16th April 2015, 01:29

Show me videolink please. Not that you measured your downloadspeed. Anyway - I'd like to see that.

touchè
geez the html info shows the download speed...

Nintendo Maniac 64

16th April 2015, 23:38

The last days the encodes were very very very slow. It needed on some videos a whole day to encode HD. Especially the days around easter.

Maybe it was because of vp9 for all videos? Could be an explanation.

All I know is, the VP9 encodes were showing up between 24-36 hours after an initial video upload, but we're hitting 96 hours now and nothing is showing up...

touchè
geez the html info shows the download speed...

The point is, download speed doesn't necessarily mean that's the actual bitrate. Since YouTube uses DASH by default, it's possible that it could be downloading multiple quality settings at once. Alternatively, when DASH is not used, it may very well try to download the video as fast as possible regardless of bitrate.

De-M-oN

16th April 2015, 23:52

It downloads only the selected file.

Nintendo Maniac 64

17th April 2015, 03:15

It downloads only the selected file.

What if the selected is "Auto"?

De-M-oN

17th April 2015, 12:22

It just switches through the files as well. It never downloads more at once.

leonccyiu

18th April 2015, 07:07

All I know is, the VP9 encodes were showing up between 24-36 hours after an initial video upload, but we're hitting 96 hours now and nothing is showing up...

The point is, download speed doesn't necessarily mean that's the actual bitrate. Since YouTube uses DASH by default, it's possible that it could be downloading multiple quality settings at once. Alternatively, when DASH is not used, it may very well try to download the video as fast as possible regardless of bitrate.

I've been wondering the same thing, I haven't seen vp9 encodes on youtube for 4 days now.

I notice the 2160p versions have a higher bit-rate than the h264 equivalent, whereas 1440p and below the bit-rate is lower. I would have thought vp9/h265 are more efficient with higher resolutions?

I am using a pentium g3258 @4.4ghz, with many chrome tabs open 1080p60 does result is quite a few dropped frames even though cpu usage isn't near max. Hardware acceleration is badly needed especially for 4k60, since the tegra x1 can decode vp9, I hope nvidia can release a 950/950ti with vp9/h265 decoding. Chrome seems to be the worse at implementing hardware accelerated decoding however.

Here is another 4k 60 video
https://www.youtube.com/watch?v=SXZ3qWAvbVs

I am using a 23 inch NEC Ips monitor which is fully 8 bit
I'll try and do more quality comparisons with h264 by having internet explorer open at the same time but I htink I do prefer the vp9 versions as in focus bright scenes do seem slightly sharper and more detailed, however regarding motion, I notice a lot of posterization which I don't recall seeing with h264, the h264 seems more consistent.

zerowalker

18th April 2015, 18:29

I have seen some for the last days, not my own videos though as that's pretty much only 0.01%.
There is no way to tell if the updated the versions or not though as there is no information in the encodings.
I mean on the h264 versions there is at least dates and some info, VP9 has nothing at all as far as i can tell.

zerowalker

18th April 2015, 19:26

I just did another Re-Encode trigger with Youtube and this time the Encoding was quite new (2014), but there was big improvements.

Here is a link: http://screenshotcomparison.com/comparison/122657

It uses less (marginal) bitrate, but it looks a lot better. Worth noting is that this is even a still position, there was no movement in the video at this spot, meaning both should behave as similar as possible.
Did not expect this kind of improvement from only a year back, i assumed the results would either be identical or marginal at best, so i am quite fascinated.
(This is NOT vp9, it's h264/avc).

Nintendo Maniac 64

18th April 2015, 23:06

.I am using a pentium g3258 @4.4ghz, with many chrome tabs open 1080p60 does result is quite a few dropped frames even though cpu usage isn't near max. Hardware acceleration is badly needed especially for 4k60, since the tegra x1 can decode vp9.
The problem is that Chrome's VP9 decoder is quite unoptimized. Download a 60fps VP9 video and play it back in the likes of MPC-HC 1.7.8, and you'll have much better performance.

For reference I too have a Pentium G3258, but I only run mine at around 3.8GHz for HTPC-like reasons. Speaking of HTPC, my HTPC's 2.4GHz Core 2 Duo can run 1080p 60fps VP9 videos in MPC-HC without any framedrops, but CPU utilization is like 90-95% (I also normally use D3D Fullscreen which saves a teeny bit of CPU).

I have seen some for the last days,

We're talking about new video uploads though, hence the hour-counts. That is, no video uploaded in the last 96 hours have any VP9 encodes (though it'll be 120 hours at this point).

zerowalker

19th April 2015, 08:41

EDIT: Forget what i wrote, the Date is different in the Channel compared to it's actually publish date.
So i haven't seen any vp9 as well.

https://www.youtube.com/watch?v=prWjk8UlRPs
Or am i thinking this wrong?

But any other video (1-2 days, even the short and popular ones are still avc).

STaRGaZeR

19th April 2015, 12:30

Still too soon to call victory, but have they ditched VP9 and instead sanitize their x264 settings?

sneaker_ger

19th April 2015, 12:33

Doubtful (about the VP9 part). Probably just working on their back end.

leonccyiu

20th April 2015, 03:56

The problem is that Chrome's VP9 decoder is quite unoptimized. Download a 60fps VP9 video and play it back in the likes of MPC-HC 1.7.8, and you'll have much better performance.

For reference I too have a Pentium G3258, but I only run mine at around 3.8GHz for HTPC-like reasons. Speaking of HTPC, my HTPC's 2.4GHz Core 2 Duo can run 1080p 60fps VP9 videos in MPC-HC without any framedrops, but CPU utilization is like 90-95% (I also normally use D3D Fullscreen which saves a teeny bit of CPU).

If you use a dynamic multiplier and adaptive voltage, then a low loads it's pretty much like stock, and you can have the additional power when you need it. Are you using the stock cooler? Admittedly I don't think going to 4.4ghz for 3.8 matters too much for decoding video, have extra cores is much more important.

Using dxva checker does your pentium g3258 have hardware accelerated h264 QFHD decode? Mine hasn't which I am disappointed about, my cpu cooler makes a mild grunt decoding 4k h264, the high bit-rate demos from demo-uhd3d result in dropped frames.

You're right google's vp9 decoder is horrible and results in stutters. Using 64 bit Daum Potplayer with ffmpeg's vp9 decoder I can smoothly play 2160p30 now in vp9 although my cpu cooler makes a mild grunt. Cpu usage occasionally touches 100 but it doesn't report many dropped frames apart from the few when I switch to full screen. Using potplayer I can enter the youtube link and select the stream without having to download, I can also use madvr.

My internet connection is only 10 mbps down, so I let the videos buffer a bit, but I have to say for the quality difference compared to 1080p is huge even on a 1080p monitor. It's a bit wasteful to be decoding 4k vp9 for 1080p output but it's like watching blu-ray at times.

I find the bitrates to be highly inconsistent for every video, and I take back what I said for 4k vp9 videos being of a higher bitrate than their h264 equivalents, it's true for some especially newer videos, but often it's the same at around 22mbps and sometimes as low as 12 for older videos.

Nintendo Maniac 64

20th April 2015, 04:19

If you use a dynamic multiplier and adaptive voltage, then a low loads it's pretty much like stock, and you can have the additional power when you need it.
I do this already; that is not the "issue". The main thing is that my overclock is just enough to run a specific CPU-intensive programs at full speed for my desired setting, and in the future I'll probably even lower my overclock once the next version of said program is released (it'll have new optimizations with an updated backend).

The whole key here is reducing heat output into the room the PC is located in rather than reducing the CPU core temperature - my CPU fan already runs at the slowest speed possible without my CPU ever reaching 60c, and I don't even have a case fan other than the PSU fan (to put things into perspective, the loudest thing in my entire PC setup is a circuitry bussing coming from my monitor).

Remember, the lower the clock, the more you can undervolt; 3.8GHz is especially a nice sweetspot for me since I can do just that and undervolt by 0.027v while staying stable, but if I go up to even only 3.9GHz I have to up the voltage to something above the stock voltage.

Are you using the stock cooler?
No.

does your pentium g3258 have hardware accelerated h264 QFHD decode? Mine hasn't which I am disappointed about

Is this a thing? I thought all Haswell iGPUs were capable of 4k h.264 DXVA hardware decoding... have you tried it via MPC-HC with its internal LAVfilters set to decode with DXVA?

Doing some research on the issue, I found your post on AVS...you claim that you're using the newest driver - I'm using the one before the newest, 15.36.14.4080, from 2015-01-12 (in the GPU control panel it says my driver version is "10.18.14.4080").

Using dxva checker does your pentium g3258 have hardware accelerated h264 QFHD decode?
I can decode 30fps 3840x2160 h.264 via DXVA in MPC-HC without dropping any frames at only around 10% CPU utilization; alternatively I can decode the exact same video but via QuickSync in MPC-HC without any dropped frames at around 40% CPU utilization.

My internet connection is only 10 mbps down
I laughed; my internet is only 3 Mbps down (and 768 Kbps up)

EDIT: Since the 4k h.264 DXVA decoding issue on Pentium G3258 is quite off-topic, we can continue that specific sub-discussion in your AVSforum thread that I mentioned earlier if you'd like (I have an AVS account already).

De-M-oN

21st April 2015, 09:34

but I have to say for the quality difference compared to 1080p is huge even on a 1080p monitor.

Yes of course. What do you expect? That you dont see the better bitrates on your monitor?

I dont get why so many people assume that they dont gain higher quality if their monitor doesnt support the higher resolution.

A video is more than just resolution and of course you see better bitrates with your monitor as well.

I find the bitrates to be highly inconsistent for every video

They aren't.

But it depends on complexity. You wont gain 20 mbit from them if your video is just a desktop tutorial video.

It looks like a combination of CRF and VBV-Maxrate.

And if the vbv-maxrate is overloaded even with their crf around 30 or whatever they use, the blocking begins - very noticable effect at 1080p. If you stay within the CRF it looks relative good, but if you break the limit and vbv-max has to brake the bitrate - then blocking starts - the more complexity - the higher the problem ^^

1920x1200 gets the 1440p quality level and has 10000 kbit. This looks quiet good. When they finally would activate HFR encode for this level the video should look even better.
Imagine 25fps bloated to HFR and that on the 1440p quality level - that should result in very good quality ;D

Thats why I sooooo damn hope that they finally activate HFR 1440p for public meh.

zerowalker

21st April 2015, 09:43

but I have to say for the quality difference compared to 1080p is huge even on a 1080p monitor.

You actually most of the time get better quality with higher resolution even if the bitrate are identical.
Meaning, 4k lossless rather than 1080p lossless on an 1080p monitor will better.
Guessing it's kinda like games with AA vs none-AA, you can get less "rounding-errors"(incorrect term) or something.

Thats why I sooooo damn hope that they finally activate HFR 1440p for public meh.

Damn and here i sat wondering why my 1920x1200 60fps never got that resolution done.
I forgot it was > 1080p that was in beta, i just assumed it was 4k for some reason. This kinda sucks for me as my recordings are "messed up" now;P

btw Demon, you had this problem: http://forums.guru3d.com/showthread.php?p=5054362#post5054362 ?

De-M-oN

21st April 2015, 09:48

yes, but they said they will as soon as it is activated give you the qualities also for already uploaded videos.

And I really wonder what takes so much time for them to enable it for public - especially considering that they already have working testvideos :/

zerowalker

21st April 2015, 10:02

Well it's always like that.
My guessing is they need to see that it works for everyone, what bitrates should they use, will those bitrate work for enough ppl around the globe.
Many variables.

But in a nutshell i am saying the same, if you got it working, publish it;P!

Nintendo Maniac 64

21st April 2015, 23:08

Don't forget about chroma resolution! That's another reason why even "lossless" 2160p would look better on a 1080p display than "lossless" 1080p would - because 99.99% of all video uses 4:2:0 chroma rather than full 4:4:4. This should also mean that "lossless" 4:2:0 2160p downscaled to 1080p should look extremely similar to lossless 4:4:4 1080p.

(the reduced chroma is why I said "lossless" in quotes, hence why I didn't use quotes in regards to 4:4:4 lossless)

zerowalker

22nd April 2015, 06:26

Indeed, didn't think that far. Half Luma would result in "Twice Chroma" if it was half of that luma.
Well not exactly but in a rough state.

Though Chroma often only looks better in Games in my experience, and it's mostly Edges and Colored Text.
Real Life has too much "Noise" that half chroma doesn't have that result.

Nintendo Maniac 64

22nd April 2015, 08:24

Just an FYI regarding there not being any new VP9 encodes, this also occured sometime last December as well and lasted for 10 or so days before new VP9 encodes started appearing again.

zerowalker

22nd April 2015, 08:31

Interesting, did anything new happen with those encodes?
As i am guessing now that they update to 1.4.0,
but then again that should probably go smoother as i think they use ffmpeg and update from time to time which doesn't give X days of nothingness.

zerowalker

28th April 2015, 06:47

Youtube seems to make vp9 videos again.

Nintendo Maniac 64

28th April 2015, 09:28

Must be really recent because it wasn't there 12 hours ago and so far it's only on recent uploads (~a day old or so) or on videos with a ton of views (read: millions).

In just a couple weeks it'll have been exactly 1 years since I uploaded my F-Zero GX test footage, so perhaps it'd be a good idea to upload a new copy and see how their VP9 encoding backend has changed since then (assuming that the change is noticable at all).

zerowalker

28th April 2015, 15:25

I have seen some about a week old with less then a million, but yeah still very popular.

You don't have to re-upload just do the trigger thing, do some enhancement then reverse.

(Though download the file first, and check the "Date Modified") so you know when the file was encoded as vp9 doesn't keep dates.

Nintendo Maniac 64

28th April 2015, 23:04

I was thinking about re-uploading so that I could keep an easier record of the differences...besides, I've already got copies of all the VP9 encodes.

zerowalker

29th April 2015, 04:58

Oh, well do remember that they re-encode from time to time anyhow.

De-M-oN

29th April 2015, 16:49

Oh, well do remember that they re-encode from time to time anyhow.
They never did reencode my videos.

Even my Doom 2 LP of 2011 has still the old encodes.

zerowalker

29th April 2015, 16:51

Hmm.. my old 2011 videos has an encoding from 2014 which doesn't make much sense.

Nintendo Maniac 64

2nd May 2015, 06:07

2015 re-upload of my F-Zero GX test video is now up:
https://www.youtube.com/watch?v=DFy6yKkV-Jg

Note however that, as of this post, there are no VP9 encodes just yet.

For reference here's the 2014 version:
https://www.youtube.com/watch?v=BLxHjixnf3c

EDIT: Well here's a surprise, the 720p DASH h.264 encoding for the 2015 version looks quite a bit better than the very same format in the 2014 version! And yet they both practically use the same bitrate.

EDIT 2: Uhhh... is it just me or does the 2015 version of the non-DASH fmt22 format look worse compared to the same format on the 2014 version? O_o

EDIT 3: Now I know it's not the best idea to compare video quality via screenshots, but I just know I'm going to get asked for them anyway (like I was previously). For reference, screenshots A and B are the same exact frames that I used for the screenshots back on page 2 of this very thread but this time I've added in a third screenshot as well.

[Source]　MP4 AVC (6169kbps): Screenshot A (http://i.minus.com/ibcUeYgI61xBig.png) ~ Screenshot B (http://i.minus.com/iVwePXc6G4oAc.png) ~ Screenshot C (http://i.minus.com/iilkuIQOechoY.png)

2014

DASH AVC (2112kbps): Screenshot A (http://i.minus.com/iboSw4RSHtp8A1.png) ~ Screenshot B (http://i.minus.com/iiTeCt5kesP36.png) ~ Screenshot C (http://i.minus.com/inacAy8QXo8Ew.png)

fmt22 AVC (2886kbps): Screenshot A (http://i.minus.com/insuN23oU8IV5.png) ~ Screenshot B (http://i.minus.com/iZ0UAYwbA8wpN.png) ~ Screenshot C (http://i.minus.com/ibnMvG7YD3Wcmp.png)

WebM VP9 (2182kbps): Screenshot A (http://i.minus.com/i9BMg79d3ReLj.png) ~ Screenshot B (http://i.minus.com/ibiPJNBePjunbh.png) ~ Screenshot C (http://i.minus.com/iyr6M7rrz8TwM.png)

2015

DASH AVC (2125kbps): Screenshot A (http://i.minus.com/ibk2PLJxjSqlLq.png) ~ Screenshot B (http://i.minus.com/ib0sNe4K5R6u8Z.png) ~ Screenshot C (http://i.minus.com/imgRW8KSOh1ch.png)

fmt22 AVC (2894kbps): Screenshot A (http://i.minus.com/iDg7MKthoJdN.png) ~ Screenshot B (http://i.minus.com/idMGsiSv4VZtJ.png) ~ Screenshot C (http://i.minus.com/igYTokwQOvX7J.png)

WebM VP9 (1686kbps): Screenshot A (http://i.minus.com/iy3GuVxJAlUXc.png) ~ Screenshot B (http://i.minus.com/ibp2LjLZjrLcKU.png) ~ Screenshot C (http://i.minus.com/iQNT1KKBfYOHt.png)

mzso

2nd May 2015, 11:42

2015 re-upload of my F-Zero GX test video is now up:
https://www.youtube.com/watch?v=DFy6yKkV-Jg

Note however that, as of this post, there are no VP9 encodes just yet.

For reference here's the 2014 version:
https://www.youtube.com/watch?v=BLxHjixnf3c

EDIT: Well here's a surprise, the 720p DASH h.264 encoding for the 2015 version looks quite a bit better than the very same format in the 2014 version! And yet they both practically use the same bitrate.

EDIT 2: Uhhh... is it just me or does the 2015 version of the non-DASH fmt22 format look worse compared to the same format on the 2014 version? O_o

EDIT 3: Now I know it's not the best idea to compare video quality via screenshots, but I just know I'm going to get asked for them anyway (like I was previously). For reference, screenshots A and B are the same exact frames that I used for the screenshots back on page 2 of this very thread but this time I've added in a third screenshot as well.

2014

AVC DASH (2153kbps): Screenshot A (http://i.minus.com/iboSw4RSHtp8A1.png) ~ Screenshot B (http://i.minus.com/iiTeCt5kesP36.png) ~ Screenshot C (http://i.minus.com/inacAy8QXo8Ew.png)

AVC fmt22 (2942kbps): Screenshot A (http://i.minus.com/insuN23oU8IV5.png) ~ Screenshot B (http://i.minus.com/iZ0UAYwbA8wpN.png) ~ Screenshot C (http://i.minus.com/ibnMvG7YD3Wcmp.png)

VP9 WebM (2131kbps): Screenshot A (http://i.minus.com/i9BMg79d3ReLj.png) ~ Screenshot B (http://i.minus.com/ibiPJNBePjunbh.png) ~ Screenshot C (http://i.minus.com/iyr6M7rrz8TwM.png)

2015

AVC DASH (2165kbps): Screenshot A (http://i.minus.com/ibk2PLJxjSqlLq.png) ~ Screenshot B (http://i.minus.com/ib0sNe4K5R6u8Z.png) ~ Screenshot C (http://i.minus.com/imgRW8KSOh1ch.png)

AVC fmt22 (2951kbps): Screenshot A (http://i.minus.com/iDg7MKthoJdN.png) ~ Screenshot B (http://i.minus.com/idMGsiSv4VZtJ.png) ~ Screenshot C (http://i.minus.com/igYTokwQOvX7J.png)

VP9 WebM: [not yet available]

Screenshots from the original would help judging the quality of the encodes and not just compare them.

Nintendo Maniac 64

2nd May 2015, 20:04

Screenshots from the original would help judging the quality of the encodes and not just compare them.

Last time I posted the original upload I was criticized that said video, and I quote, "seriously lacks bitrate" even though I had stated that I was aware that the encode was older and had not-so-great settings; for reference it was one of if not the first lossy encoding via x264 that I had ever done.

I'll probably make a better source encode at some point in the future seeing how the original recording was 1080p (and in lossless too!), but I felt that I needed to at least upload the original encode for comparison reasons.

Also, even this mere 4-minute not-so-great 720p 30fps encode (which is 200MB) took 50 minutes to upload...yay for my DSL only having ~75KB/s upload.

Nevertheless, in the YouTube video descroption(s), you can find a link to the original uploaded MP4. For reference, said MP4 does still look considerably better than the YouTube encodes, but remember - this is a video of a 12 year old GameCube game, so any textures or the like won't be that high of a resolution.

zerowalker

3rd May 2015, 10:42

I looked and compared the old and new vp9 versions.
If i am not doing something wrong, the new was is (for most part) quite a lot better in quality.
It's really noticable when there is small detail, old one is a blurfest compared to the new one in most cases.

EDIT: Question: Was the video originally a lossless upload?

mzso

3rd May 2015, 13:12

EDIT: Question: Was the video originally a lossless upload?

He told above you that it was not a good quality encode, certainly not lossless

zerowalker

3rd May 2015, 14:32

Oh, my bad;P

Nintendo Maniac 64

3rd May 2015, 20:37

The VP9 encodes have come in now (they weren't there 12 hours ago) so I'll get right on that. And since you guys are quite aware now that the original upload was not lossless, I'll probably stick in some screenshots from that and maybe even add in some actual screenshots from the lossless recording as well.

He told above you that it was not a good quality encode, certainly not lossless

Well a few of the issues are present in the game itself and the original lossless recording - there's still some aliasing, particularly in the parts of the "mesh grid", and the road textures "fade in" as they get closer to the camera (likely to save on texture memory - not noticable in 480p but it can be when rendered in HD)

EDIT: Screenshots!

[2015]　WebM VP9: Screenshot A (http://i.minus.com/iy3GuVxJAlUXc.png) ~ Screenshot B (http://i.minus.com/ibp2LjLZjrLcKU.png) ~ Screenshot C (http://i.minus.com/iQNT1KKBfYOHt.png)

EDIT 2: Well from the screenshots alone the quality really doesn't seem all that different from the 2014 encode...only Screenshot A really looks distinctly better on the 2015 version, and even then I wouldn't really say it's by that much.

EDIT 3: mzso, I would greatly appreciate it if you didn't quote my post in post #346 that you made - that way I can edit and update my post and the like without there being an non-updated version quoted directly below it...

EDIT 4: Holy crap, no wonder the 2015 VP9 encode doesn't look much different - the bitrate is only 1686kbps! Wow, there's definitely been some quality improvements if they can reduce the bitrate by 25% and still have about the same quality. Man, if only I could have tested 1080p...well maybe I could, but it would have to be a really short video - like only 30 seconds at most.

EDIT 5: Remember, it wasn't the greatest encoding...

[Source]　MP4 AVC (6169kbps): Screenshot A (http://i.minus.com/ibcUeYgI61xBig.png) ~ Screenshot B (http://i.minus.com/iVwePXc6G4oAc.png) ~ Screenshot C (http://i.minus.com/iilkuIQOechoY.png)

De-M-oN

3rd May 2015, 22:08

I hate that. Why dont they just treat (I hope its the right word for "gönnen" xD ) us the better quality? ._.

Thats the general problem .. if there is more efficiency they react with reducing the bitrate instead of treating us the better quality.. And this worries me - because this is then a deadend. A black hole. We will never see better quality if they react like this to new codecs.

Gladly 1440p @ 1920x1200 looks okay at least. Especially it would with VP9. At least there it is that 30fps gets about 10000 kbit @ h.264 and at HFR it gets 10000 kbit VP9 - so even better quality then. Damn I cant wait anymore for 1440 HFR to be activated for all..

Nintendo Maniac 64

3rd May 2015, 23:22

I hate that. Why dont they just treat (I hope its the right word for "gönnen" xD ) us the better quality? ._.

Maybe this will answer some questions...
http://youtube-eng.blogspot.com/2015/04/vp9-faster-better-buffer-free-youtube.html

tl;dr - more bitrate isn't useful if most people's connections can't handle it.

if there is more efficiency they react with reducing the bitrate instead of treating us the better quality.. And this worries me - because this is then a deadend. A black hole. We will never see better quality if they react like this to new codecs.

A large part of this is because they're trying to make YouTube more accessable to everyone in the world. Heck, I live in northeast Ohio and I only have a 3Mbps down connection! Personally it looks to me like they're trying to make 720p be a new "baseline" since there are situations (particularly those involving text) where 720p compared to, say, 360p, is quite beneficial even at a low bitrate. From there the higher quality formats (1080p+) are arguably more a case of "gravy" in that there are very few things that absolutely need 1080 vertical pixels.

Now to clarify, nothing actually needs even 360 vertical pixels - my point is that there are situations where the content in question was designed with the expectation of an HD display to the point that even 480p at lossless quality is worse than 720p at 1.5Mbps.

De-M-oN

3rd May 2015, 23:30

no bandwidth would be increased if using the same bitrates. But reducing it further is just annoying. Sure this helps lower end internet connections - but if you select 1080p or higher - it should be decent quality then. These are not quality levels with intention for low end internet anyway. And 1080p with only 4000 kbit is already low enough anyway. Why reducing it further with vp9? : /

4000 kbit can watch a DSL6000 user - thats already fair enough imo if a DSL 6k user can watch even 1080p. Why reducing 1080p further? Just pointless ..

Nintendo Maniac 64

3rd May 2015, 23:36

1. You posted before I finished editing my own post.

2. I haven't been testing 1080p because I cannot stream it, so I have minimal opinion on it. Again, I only have a 3Mbps DSL connection.

zerowalker

8th May 2015, 14:26

Question, doesn't .Webm support Opus?
Thought it was "made" for VP9 + Opus, or well that was the idea or something.
But according to ffmpeg opus isn't valid for it.

nevcairiel

8th May 2015, 15:26

webm is VP8/VP9+Vorbis/Opus.
A quick check in the ffmpeg sources also suggests that it allows all these formats. Make sure your version is up to date.

zerowalker

8th May 2015, 15:36

Hmm, i got this guys builds: http://ffmpeg.zeranoe.com/builds/

But do you mean is limited to VP9+Opus or something?
Or can it be any supported video with any supported Audio (like MKV, as i am pretty sure Webm is like a branch of it or something).

I can't combine libx264 with libopus.

nevcairiel

8th May 2015, 15:49

webm does not allow H.264, so using that will not work. It needs to be VP8 or VP9 for Video, and Vorbis or Opus for audio. Nothing else is valid in webm.
If you want a MKV, then name it MKV. That'll allow all codecs.

zerowalker

8th May 2015, 15:53

Ah so that was the culprit, thought it was Opus.
Thanks for the explanation:)

zerowalker

9th May 2015, 11:21

foxyshadis

9th May 2015, 18:44

Youtube seems to be spamming out VP9 now.
Previously only very rare videos got VP9 for me, the popular ones.
And no private became VP9 at all.

Now i noticed that some videos (from this week etc) become VP9 no matter of popularity, even private ones.
My guess is that they have got 1.4 up and running and as the speed is better they can no produce more videos and don't have to be too picky.

That's fantastic news. Have you noticed them switching out already-encoded versions for better ones yet? I hope that starts happening soon, because I doubt most uploaders would ever intentionally trigger a re-encode.

Skarstorm

9th May 2015, 19:19

Is there a link between a video that is being cached over multiple servers around the world and turning the video from video/mp4; codecs="avc1.4d401f" to video/mp4; codecs="VP9"
and when the view count is stuck on 301+ views, this refers to recent (let's say the first +/-24h) uploaded video's
https://www.youtube.com/watch?v=oIkhgagvrjI&feature=youtu.be&t=93

zerowalker

9th May 2015, 19:43

I have no idea which version of VP9 they use as there is no way to tell. They don't store any information and the file.
But something must have changed, i have never gotten VP9 on 0 views and even private ones.

Skarstorm

9th May 2015, 20:14

So could their be a relation between changing to VP9 and viewcount =(caching over multiple server when video starts to become popular)
OR between changing to VP9 and the time =(encoding to VP9 takes a hell of a lot more time then to AVC, this is with the assumption that YouTube alsway recodes the uploaded video's)

zerowalker

9th May 2015, 20:30

Well if videos that are private with 0 views gets encoded to VP9 their system of "Viewcount" has ended.
Then again i think some popular channels get VP9 no matter if there Videos are Popular or not (but the Channel must be popular to obtain this).
My channel can't rly be called popular so that can't be it. So i think they just updated their systems and VP9 encodes are now faster which allows then to use them more.

Garteal

10th May 2015, 23:35

When I uploaded some test videos some days ago, my videos also got encoded to VP9 within an hour or two.
They were test videos thus set to unlisted so it doesn't have anything to do with view count. I'm not popular or anything either so I doubt that's it.
I think they might've made it publicly available and are maybe applying the changes in waves or something if you aren't getting it.

zerowalker

12th May 2015, 04:33

That's great to hear:)

After checking more videos after i uploaded a lot, i can say that it's completely random.
I got two versions of same videos (different audio tracks) one is private other public.
And usually the private seems to become VP9 while the public doesn't (but think it's just random, sometimes none gets it, and triggering a re-encode doesn't seem to have much affect).

I do however think that you can force trigger VP9 if you upload with the Youtube API at certain settings.
Sadly i don't want to use it as it doesn't keep original file names (Raw file), it just says "Unknown".
But it was some tests i did before where i uploaded through Chrome and Youtube API,
but it might just have been that it just happened to be lucky even though i did several tests.

De-M-oN

12th May 2015, 13:04

http://www.letsplays.de/forum/index.php/Thread/22798-Simple-Java-Youtube-Uploader-3-2/

Since this isnt the webuploader, it must use the API too I would say - and if I upload with it - the filename is listed at youtube.

But this doesnt ensure vp9 encodes as well. Its completely random.

zerowalker

12th May 2015, 16:08

Maybe.. i think though the API is locked to your account?
My application doesn't work for any other account as far as i know, they are very strict.
And you you have to do a certain setting. But will do more tests on it later just for fun, like 3 short videos or something to see if 100% works on API or if it's random.

zerowalker

12th May 2015, 18:32

Okay did a test.

5 videos through Chrome = 4/5 became VP9
5 videos with Youtube API v3 with certain settings = 3/5 became VP9.

So conclusion is that i was just very lucky before when i did tests with it.
It's completely random.

Nintendo Maniac 64

12th May 2015, 23:06

We should clarify that it's random in regards to non-popular channels that don't have videos with high view counts. If you have even 1 video that has like over 100,000 views regardless of channel popularity or subscriber count, then it's extremely likely that you'll be getting VP9 encodes.

zerowalker

13th May 2015, 00:11

I got 2 videos with over 150k views, that hasn't really helped me;P
If you meant that it would make the channel "popular" and so get you VP9 encodes.

But of course those videos are in VP9.

Nintendo Maniac 64

13th May 2015, 23:05

I got 2 videos with over 150k views, that hasn't really helped me ;P.

Maybe it takes several videos? (as in a few more than 2)

zerowalker

13th May 2015, 23:30

Probably, i only got 2 videos at 150k+ and then 65k, 40k, 36k and 32k. (then it's 16k and it get's much ruther down).
So, i would guess anyone who usually have 150k would have a much more wider range of those kind of videos, i just have "some that happened to be popular" kind of videos.
Well it least it proves that having a few popular videos doesn't get you into the "Popular Channel" list;P

On another note, i did a Re-Encode of a "half popular" video some time ago which actually got VP9 after that.
So it may be that when it Encodes it does some checks, and if it's met it will become VP9 (at least before, now it's probably random || popular).
I also tried Re-Encoding a short clip several time to see if i can force the Random VP9 Encode, but i doesn't seem to have much effect, so very confused to how this all works, can't be completely random.

Nintendo Maniac 64

14th May 2015, 01:31

All I know is, I have videos (really still images with music) that are 4 or so years old and they have VP9 encodes... but they also have 100 000+ views. And to clarify, I never re-encoded them or did any other kind of video editing and/or post-upload processing.

zerowalker

14th May 2015, 07:15

Yeah can say same. Those that are 100k+ are very old, think from 2011.
Never re-encoded them either, they just became VP9 some time after VP9 got activated i think.
So very popular videos clearly are tagged for "immediate encoding".

zerowalker

20th May 2015, 10:04

Is it just me or is the quality on both AVC and VP9 awful right now compared to some week ago?
I mean my videos are Really blocky, even though the bitrate is like it used to be as far as i can tell.

EDIT: Okay it's even more Bitrate for much less quality.

Not these pictures aren't the best for comparison, but the blockiness rly is that bad everywhere, no matter if it's bright or dark.

//The dates are different. First one is Year-Month-Day, other one is Year-Day-Month. (So first is 19 May other is 10 May).
Video bit rate : 9 236 Kbps

http://ladda-upp.se/files/2015/b149186.png

Video bit rate : 6 798 Kbps

http://abload.de/img/youtubeworsequality4zumz.png

De-M-oN

20th May 2015, 18:19

JPG Compression wasnt a good idea for image comparison. Use PNG.
_
First one is very bad compressable with the trees, grass etc - such vegetation is fine detail and offers no similarities especially while movement.
2nd one is the complete opposite - very solid textures

zerowalker

20th May 2015, 19:51

It's png, guess it's the site.

Done some more tests, not exactly sure if it's random or if i am just getting to picky on Youtube.
I know this has happened before at a time. Everything became blocky instead of smudged.
Might just be that i just noticed it and go like, "Wait a minute?", just not seen this with VP9;P

Nintendo Maniac 64

20th May 2015, 23:13

I could make a second reupload of my F-Zero GX footage and compare it to copy I just uploaded a few weeks ago.

It's png, guess it's the site.

It may have been uploaded as PNG but it sure got converted to JPEG - I even downloaded it and checked the file contents via XnView (which will report the proper format regardless of file extension)

My theory is that, without having an Imgur account, any image over 1MB will be recompressed. I don't have an Imgur account myself since I normally use Minus, so I can't say for sure (I can say though that Minus preserves large-filesize PNGs)

De-M-oN

20th May 2015, 23:23

Why using such bad image hoster then?

http://www.abload.de/

No conversion here.

zerowalker

21st May 2015, 09:49

Nintendo Maniac 64, please do, i am probably wrong but you never know.)

And yeah it's Jpeg, it converted.

I usually use Tinypic but then everyone complains and tell me to use Imgur,
but if they convert to jpeg then it's crap;S

Will see if i can find something, else i will use the German site you provided Demon:)

dapperdan

21st May 2015, 14:29

Interesting comment from what appears to be a Youtube employee:

https://www.reddit.com/r/youtube/comments/36cnil/what_determines_if_an_uploaded_video_will_be/crdnsbj

He notes that due to A/B testing, some people may see different behaviour than other users, and that you can't assume that two similar users will experience the same thing, at least not before it rolls out widely.

zerowalker

23rd May 2015, 09:02

Guess that's what they did.
Currently none of my latest videos are VP9, before pretty much ever other video became vp9.
Either that or i just got bad luck;P

NikosD

27th May 2015, 10:04

A lot of changes to YouTube services.

YouTube released 60 frames per second for live streaming.

The feature launch as a preview on HTML5-supported browsers. When a user starts a stream, YouTube automatically encodes it at 60 frames per second on 720p and 1080p.

In addition to adding 60 frames per second on live streams, YouTube also enabled HTML5 playback so that you can watch a previous section of the screen at any time and then fast-forward at 1.5x or 2x to catch up to the live broadcast.

De-M-oN

27th May 2015, 17:33

And my video got the HFR encode for 1440p (1920x1200) :)

http://abload.de/img/desktop_2015_05_26_23c2s5b.png

https://www.youtube.com/watch?v=bBoBhj8A0xI

And it looks awesome !

zerowalker

27th May 2015, 18:43

Damn forgot that HFR at 1080p+ only happens with VP9.
Then none of my 1920x1200@60fps videos will become that -_-.

Nintendo Maniac 64

27th May 2015, 23:13

Then none of my 1920x1200@60fps videos will become that -_-.

Who knows, maybe because it's only available for VP9 that it will work as a guaranteed way to trigger the VP9 encodes? However it may only work for new uploads...

zerowalker

28th May 2015, 00:18

Who knows, maybe because it's only available for VP9 that it will work as a guaranteed way to trigger the VP9 encodes? However it may only work for new uploads...

Doubt it, but maybe. I did try to trigger re-encode but that didn't change anything, still 1080p60.

But i wonder, is 4k60 rly out of Beta, i haven't heard or read anything about it. Didn't they make a huge deal of it when 60fps went live?

De-M-oN

28th May 2015, 00:21

4k60 is ID 315 and exists as well: https://www.youtube.com/watch?v=lBtUWHCacI4

Youtube said that - when HFR 1440 and 4k is activated for all - earlier videos will get the missing encodes as well.
And I believe them - considering that a 59,94 fps video of year 2011 got HFR 1080p now lol.

zerowalker

28th May 2015, 03:41

Tried uploading a 1920x1200@60 to youtube and it's only in 1080p60 (very short clip).
No VP9 of course.

My guess is that they will have to "fix" their system so that anything above 1080p60 will have to be VP9 encoded.
So even if the video is not flagged for VP9, 1080p60+ will be VP9.

De-M-oN

28th May 2015, 16:18

Better Solution: VP9 for all and every video ..

zerowalker

28th May 2015, 23:00

Better Solution: VP9 for all and every video ..

Would love that, but guess it's not a realistic option.

Also i did a test.
I took a 1920x1200@60 video that was VP9 from before and re-triggered it.

It re-encoded and became VP9 again after awhile (which means that a video flagged as VP9 is Always VP9), cause it was AVC for awhile during this.

But importantly, it never got 1920x1200@60 like before, it only goes to 1080p60, even though it's VP9.

EDIT:

Did some tests, uploaded a file privately and triggered encodes several times.
It became vp9 at a point, but when i triggered again it was gone.
So the "Flagged to VP9" becomes invalid. (Also it didn't become 1920x1200@60, still triggering from time to time to see if that changes).

zerowalker

5th June 2015, 15:24

Was able with a test clip after several attempts or re-encoding to a while ago make it 1920x1200@60.
I then tried it with another clip i had and re-encoded that to no avail.

So re-encoding CAN trigger it, but it's completely random, as is VP9.
Sometimes it gets VP9, sometimes it doesn't. (I think having views will probably force it though).
60fps seems to be the same, it CAN happen but much more rare, and views will most likely increase that a lot.

According to Youtube 4K@60 isn't "Active" yet if you read it here: https://www.youtube.com/testtube
And if playing 4K videos at 3840x2160 isn't enough to melt your device, try doing it at 60 frames per second. We've combined these two cutting-edge features for a few videos in this playlist.
So i am guessing they are just "playing around", let's hope they activate it on everything:)

EDIT:

Worth noting is that i actually got Opus audio files during 1 or 2 re-encodes for a video.

Nintendo Maniac 64

5th June 2015, 23:15

I must ask, what are you using to determine Opus audio? My primary method does not support Opus audio yet...

zerowalker

5th June 2015, 23:20

I must ask, what are you using to determine Opus audio? My primary method does not support Opus audio yet...

Youtube-dl.

Sadly it seems that audio (at least when played on youtube) on a certain video having that is delayed like -50 to -100 ms or something.
Might not be opus playing as i have no way of telling through the Youtube Player (HTML5).

Here is my example which i finally got working to 1920x1200@60 after just now (not the test video):
https://youtu.be/_D28_MkvF74

You can see that audio is a bit off track at 0:35 when i throw down the box, the smash sound comes a bit before. (Very picky on this stuff;P).

Anyhow, this video also got Opus, 3 different tracks i think.

De-M-oN

5th June 2015, 23:37

Might not be opus playing as i have no way of telling through the Youtube Player (HTML5).

Sure you have :)
Rightclick in video and open stats view.
Look at the ID number. It has ID 308 for video and ID 251 for audio. And 251 is Opus @ 160 kbit.

And why tune @ film? Why not the default? Any reason? :)

zerowalker

5th June 2015, 23:45

Sure you have :)
Rightclick in video and open stats view.
Look at the ID number. It has ID 308 for video and ID 251 for audio. And 251 is Opus @ 160 kbit.

Ah didn't think about that, thanks;D

Damn though, then it's probably opus to blame for the desync, Youtube have had similar issues a long time ago, but that was primarily with certain audio codecs like Flac uploads.

Though again i think i have seen similar stuff recently though haven't really looked into it, might be on all uploads.

Can at least confirm that Opus 160 with 1920x1200@60 video muxed and played on my PC is in sync. So it's the damn DASH, HTML5 Player or whatnot;P

zerowalker

6th June 2015, 21:28

Sure you have :)
And why tune @ film? Why not the default? Any reason? :)

Missed that.

Well simply because it gives better picture, but i can pretty much get same results by lowering CRF. But Default normally blurs too much, and size isn't really an issue for me:)

NikosD

12th June 2015, 07:57

First 8K video appeared in YouTube, not supported by all browsers (IE)

You can check it out here with Chrome:
https://www.youtube.com/watch?v=sLprVF6d7Ug

It's a scaled up 6K video, because only a few cameras exist supporting 8K resolution.

mzso

12th June 2015, 11:31

First 8K video appeared in YouTube, not supported by all browsers (IE)

You can check it out here with Chrome:
https://www.youtube.com/watch?v=sLprVF6d7Ug

Marvellous... The don't even provide enough bitrate for 720p/1080p videos.

It's a scaled up 6K video, because only a few cameras exist supporting 8K resolution.

I wonder why is that. If I were a Hollywood studio I would have ordered than used 8k cameras for a couple years now. To make movies somewhat future proof. Sure it's at least four times the bitrate and data, but it's not like they can't afford it.

sneaker_ger

12th June 2015, 15:32

The 8K source seems to have frames missing so be careful when trying to judge decoding. And it's H.264, not VP9.

Nintendo Maniac 64

12th June 2015, 23:09

FYI that 8k video uses the same FMT as "Original" (FMT138) did before it got renamed to 2160p 4k and replaced by a different FMT (FMT266). Not only that, but FMT138 actually predates YouTube's use of VP9 altogether by something like 6 months.

mzso

13th June 2015, 10:28

benwaggoner

13th June 2015, 18:03

The quality doesn't seem very good when zoomed to 100% (meaning cropped to fullHD). It has nine times the bitrate (20.7 Mb instead of 2307Kb), I guess it should be closer to 16 times. Also the fullHD bitrate is insufficient to begin with...
If YT had something decent for FullHD like 5Mbit that would translate to 80Mbit for 8kUHD, so four times as much.
The classic rule for estimating bitrate changes required for equivalent quality is to take the change in area to 3/4ths power. So, 16^0.75=8. The principal is that each pixel has less detail as each one incorporates a smaller part of the original image, and that the pixels on the display get smaller as well.

If they're doing 9x, that seems pretty reasonable. I don't know that it's fair to do a 200% zoom. Presumably those with a UHD monitor would see the UHD stream.

That said, I don't have any idea what the use case for a 8K stream is today :). There are a few monitors >3840 pixels wide, but no one is going to be able to see the extra detail for content with a hint of motion blur.

With standard 24p stuff with a 1/48th of a second shutter, a pan that took 90 seconds to move across one screen width would be look identical at 2160p and 4320p.

(to check my math 3840/48=90, so that there would be one pixel wide horizontal blur for a 90 second pan).

Not even Ken Burns has ever done a pan THAT slow :).

A fixed tripod without wind or vibration with a very fast shutter shooting static objects is where being able to capture 8K detail could come in. But even CCD noise is going to limit actual signal detail.

Under hypothetically best conditions (fresh bulb delivering full IMAX spec brightness, perfect non-perf screen, black walls of the theater audience wearing black, exit signs at legal minimum brightness, future theoretically optimized lenses), a "4K" theatrical presentation could maybe resolve 1440p detail someday. Today, being able to see full 1080p detail is going to be a stretch even for IMAX.

mzso

14th June 2015, 10:57

Does this video's vp9 stream seem particulary wretched compared to AVC? https://www.youtube.com/watch?v=XlsIPwSj6Q0

The classic rule for estimating bitrate changes required for equivalent quality is to take the change in area to 3/4ths power. So, 16^0.75=8. The principal is that each pixel has less detail as each one incorporates a smaller part of the original image, and that the pixels on the display get smaller as well.
Okay. Though if you calculate from something decent like 5Mbit/s you get 40Mb/s which is still double of what they have.

With standard 24p stuff with a 1/48th of a second shutter, a pan that took 90 seconds to move across one screen width would be look identical at 2160p and 4320p.

(to check my math 3840/48=90, so that there would be one pixel wide horizontal blur for a 90 second pan).

Not even Ken Burns has ever done a pan THAT slow :).

A fixed tripod without wind or vibration with a very fast shutter shooting static objects is where being able to capture 8K detail could come in. But even CCD noise is going to limit actual signal detail.

Under hypothetically best conditions (fresh bulb delivering full IMAX spec brightness, perfect non-perf screen, black walls of the theater audience wearing black, exit signs at legal minimum brightness, future theoretically optimized lenses), a "4K" theatrical presentation could maybe resolve 1440p detail someday. Today, being able to see full 1080p detail is going to be a stretch even for IMAX.

It feels like you're ignoring things. By this logic with a moderately fast pan SD would look like 8k which I doubt it does.
Anyway. In my opinion framerates should be beefed up before going to 8k. 120-150 should be good. It would add way more perceivable detail.

Nintendo Maniac 64

14th June 2015, 23:29

Does this video's vp9 stream seem particulary wretched compared to AVC? https://www.youtube.com/watch?v=XlsIPwSj6Q0

The source content is interlaced and has not been de-interlaced at all. If you look at the 720 resolutions rather than the 1080p resolutions, you'll see that the quality of VP9 vs AVC is as you would normally expect, likely because the downscaling blends together the interlaced lines effect.

It may be safe to say that VP9 is not optimized at all for this type of interlaced content, especially considering how much more recent VP9 as a format is compared to AVC.

benwaggoner

14th June 2015, 23:40

It feels like you're ignoring things. By this logic with a moderately fast pan SD would look like 8k which I doubt it does.

Anyway. In my opinion framerates should be beefed up before going to 8k. 120-150 should be good. It would add way more perceivable detail.

More like with a moderately fast pan, everything looks SD.

>60p video hasn't been shown to be all that visible with real world content, but the jump from 24 to 60 is HUGE.

Unfortunately it is much cheaper to add more pixels than fast pixels...

-Ben Waggoner (via TapaTalk)

foxyshadis

15th June 2015, 18:55

benwaggoner

15th June 2015, 20:18

Oh geez, 8K in h264 -- at that bitrate every transform block is probably practically a solid color, that can't be good for efficiency. At least VP9 supports large enough transforms to make some sense of all that extra area, if it ever becomes fast enough to watch 8K in realtime.
Yeah, H.264 definitely shows its age >1080p. i'd expect a refined VP9 encoder to beat x264 at 2160+

(Not to mention 8K is a technology desperately searching for a market, at least until wall-to-wall screens become a thing.)
It think 8K displays will be great! Scenarios include:

Digital Signage
CAD
Collaborative touchscreens
Being able to edit two 4K videos side by side

That said, to be able to resolve at the pixel level, you'd actually need to move your chair, not just your head.

I've got a 32" UHD monitor, and find about 27" the farthest comfortable distance to read text and such. In 8K that's be 64". So, yeah, I'd need to slide by chair back and forth to see what's on the far left and far right.

pieter3d

15th June 2015, 21:17

Compression tools for 8k will need to have focus on screen content, lots of 8k content will be synthetic (rendered stuff).

Nintendo Maniac 64

15th June 2015, 23:11

No HMD VR fans in this thread? :p That's a use-case where even 16k is arguably beneficial.

benwaggoner

16th June 2015, 00:00

No HMD VR fans in this thread? :p That's a use-case where even 16k is arguably beneficial.

Although I think that definitely would be encoded as tiles, maybe even separate streams.

I did a spherical projection VR thing back in 1997. Amazing how everything old becomes new again.

mzso

16th June 2015, 11:13

No HMD VR fans in this thread? :p That's a use-case where even 16k is arguably beneficial.

I don't think that will ever happen. It's interactive CGI, can't be done with a simple video stream.

Maybe with a fixed viewpoint it would be possible. But you'd need really compact wide angle cameras very close together. Or pre-rendered CGI which would probably become obsolete pretty quickly as realtime rendering improves. With real-time rendering using eye tracking you can also exploit the uneven resolution of the eye: https://www.kickstarter.com/projects/fove/fove-the-worlds-first-eye-tracking-virtual-reality

dapperdan

16th June 2015, 12:52

I think Google announced they were going to do it with a bunch of goPros and some clever software at their last big shindig:

https://www.google.com/get/cardboard/jump/

De-M-oN

17th June 2015, 19:05

Rollout of HFR 1440 / 4k is there since 10th June.

Encode 308 looks very good!

https://www.youtube.com/watch?v=NJaRXAH4kxE

It has 1920x1200 and got 12 mbit bitrate for video.

ROTT is not easy to compress and looks bad on their 303 encode (1080p)

But Encode 308 really looks quite nice!

And with lesser complex videos it is even like the source video:

https://www.youtube.com/watch?v=JIV6Xo-MaEc

RRD

18th June 2015, 08:30

I found a way how to generate VP9/opus encodes. Go to Video Manager, pick a video click Edit -> Enchantments and then Save and wait aprox. 30 minutes. That's it.

You could use the exact same trick several months ago when YouTube implemented 60fps publicly but had NOT started re-encoding older videos; it would force YouTube to create a 60fps encode of 60fps videos uploaded before the change. So, I guess this method will always allow you to get the latest encoding process even for old videos (since YouTube keeps original files).

foxyshadis

18th June 2015, 10:48

Curious, when did they actually start keeping originals and re-encoding from them? They certainly didn't for the old 240p and 360p vids.

dapperdan

18th June 2015, 17:42

Maybe those vids were that res in the original upload? I think this one is the first video uploaded, and it's got VP9 now, but it's only 240p:

https://www.youtube.com/watch?v=jNQXAC9IVRw

chummy

2nd July 2015, 23:31

When they started 60fps i do that trick to force to reencode my old videos which i uploaded at 60fps and worked fine.

Now i wanted to force they reencode a nov/2014 1440p/50fps video which they only encoded with 1080p/50fps max option to display. But the trick dont worked this time for this and keep samething with only h264 version of the video and no VP9 alternatives.

I say that because for what i see, now they are encoding 1440p/2160p/50/60fps only in VP9 for chrome. H.264 for something 50/60fps only at 1080p or below.

De-M-oN

3rd July 2015, 02:07

Their VP9 encodes look much better too than h.264.

Nintendo Maniac 64

7th July 2015, 18:59

Finally uploded a lossless 1080p 30fps version of my F-Zero GX in Dolphin video and sure enough it has VP9 encodes; I'll type up a more in-depth post later at some point in the future.

https://www.youtube.com/watch?v=BEbIkAOCm6A

De-M-oN

7th July 2015, 21:01

Upload in at least 41 fps and in 2048x1152:

Then you get Encode 308 and it looks awesome :)

Encode 308 ist much much better than Encode ID 248 which your video got with 1080p30.

https://www.youtube.com/watch?v=KvT-CTY_yFg
or this:
https://www.youtube.com/watch?v=I8n60s_xW5s

Nintendo Maniac 64

8th July 2015, 05:06

Upload in at least 41 fps and in 2048x1152:

Then you get Encode 308 and it looks awesome :)

1. I am well aware of this because it was me that confirmed that 1152p was the minimum resolution required to achieve such a thing.

2. The raw source recorded video footage was 1080p, so I'm not going to upscale it when I could just instead re-record said replay at a natively-rendered 2048x1152 (and maybe even 60fps if I'm lucky).

3. As the whole point of me uploading that video was to do quality comparisons rather than to achieve the best quality, I would feel the need to upload a native 1080p video so as to compare to a future 1080p YouTube encoding for a native 1152p video.

4. The only reason I was even able to upload the original lossless 1080p video was because I was staying at a hotel for the last few days with quite good upload bandwidth. For reference, this is something I thought about last year when I was at the very same hotel for a similar real-life reason, but I had not brought a copy of said lossless video with me. Therefore, in the end, it was mainly about making up for previously-missed opportunities.

EDIT: Though now that I think of it, it would have been better to test an upscaled version and a native version. Of course, I would be too OCD to want to use a normal upscaler (lanczos or bicubic) and would want to use a super-fancy algorithm that can't even be done in real-time and requires me to export the video as individual frames - all of which would be quite painful to do on my laptop without at least help from my desktop.

De-M-oN

8th July 2015, 17:46

2. The raw source recorded video footage was 1080p, so I'm not going to upscale it when I could just instead re-record said replay at a natively-rendered 2048x1152 (and maybe even 60fps if I'm lucky).

Maybe it is even better compressable when you upscale it instead of native.

The best compressable scaler is Spline100 because it is a soft scaling method - but has big advantage in compression and avoids ringing artefacts too - and its just slightly less sharp, especially with that small upscale step.

Ely

11th July 2015, 13:26

I find it a bit stupid that they reencode very old 240p videos with VP9.. The results look like absolute crap. The originals weren't great either, but at least they didn't have obvious blocking like this one : https://www.youtube.com/watch?v=Ofn2A1p13Sg

foxyshadis

19th October 2015, 07:16

De-M-oN

20th October 2015, 01:13

Just noticed that Youtube has added Opus to its list of codecs, at 128kbps (though it's labeled 160kbps in my downloader). Whoa! That's a fantastically excellent audio stream, easily rivaling the AAC 192kbps. Other Opus streams are 70kbps and 50kbps.

You noticed it "early" ;D

Opus is so long in usage now lol. At least since 1440 HFR was there - and I'm sure it was there even earlier already.

Your downloader says 160 kbit, because ID #251 was at 160 kbit before, but they changed it to 128 later.

If you get AAC or Opus depends on random.

Nintendo Maniac 64

20th October 2015, 03:14

If you get AAC or Opus depends on random.

AAC goes with the AVC encodes while Opus goes with VP9 encodes, and all WebM encodes also have Vorbis as a fallback. Videos that got VP9 encodes before Opus don't have Opus; sometimes even a new video will not get Opus even if the videos uploaded before and after it have Opus and lower view-counts.

easily rivaling the AAC 192kbps

It must be noted that this is only available for the non-DASH 720p AVC encode which also maxes out at 30fps. Admittedly though it's higher quality than even 1080p AVC, though not 720p VP9.

foxyshadis

20th October 2015, 12:49

Huh, I've never seen it mentioned here, on YT Blogs, or on the Opus mailing list... obviously, it's not big news.

It must be noted that this is only available for the non-DASH 720p AVC encode which also maxes out at 30fps. Admittedly though it's higher quality than even 1080p AVC, though not 720p VP9.

And this is why Youtube gives me a migraine.

MoSal

20th October 2015, 13:31

Huh, I've never seen it mentioned here, on YT Blogs, or on the Opus mailing list... obviously, it's not big news.

Wikipedia's YouTube page is a good reference for updated formats.
Format 251 was added to the page in June 2. (https://en.wikipedia.org/w/index.php?title=YouTube&oldid=665143546)

Or you can just run:

youtube-dl -F <popular_new_video_link>

If I'm not mistaken, youtube-dl will list all available formats including the ones that are still unknown.

De-M-oN

20th October 2015, 19:54

It must be noted that this is only available for the non-DASH 720p AVC encode which also maxes out at 30fps. Admittedly though it's higher quality than even 1080p AVC, though not 720p VP9.

And this is why Youtube gives me a migraine.

I can heal you.

Because this is wrong. My 2560x1600 Video got Opus at 1440p.

If you didnt know about Opus, you maybe also didnt know, that Youtube supports 8k.

Nintendo Maniac 64

21st October 2015, 01:08

Because this is wrong. My 2560x1600 Video got Opus at 1440p.

...this statement confuses me. What is wrong about what I said, and what does it have to do with Opus being available on 1440p videos?

De-M-oN

21st October 2015, 01:34

ah now I got it. You meant the 192 kbit AAC is only available at non-dash - okay - thats correct then.

I thought you meant Opus :P

Nintendo Maniac 64

22nd October 2015, 22:10

From my own recent upload, it seems that the 128kbps bitrate for Opus is more like a target minimum average in that, according to foobar2000, it rarely goes below 128kbps but commonly hangs around 135-140kbps and even hits 160+kbps more than I'd expect; it's to the point that MediaInfo claims that the average bitrate is actually 139kbps.

Here's the "video" in question (if you count a still image with music as a video):
https://www.youtube.com/watch?v=EWErRrb6kQU

Also, interestingly enough, YouTube supports really low video framerates - that "video" in question was actually only two frames total that played at a framerate of 0.0011fps. While YouTube's own encodings seem to have a minimum framerate of 6fps, doing this does allow you to reduce the filesize for upload with cases like this where it's a single image for several tens of minutes.

foxyshadis

23rd October 2015, 01:30

Oh god, I can't stop listening.

That video brings up some very strange frame-sizes on the downloader; Youtube must give the actual size in the streaming links now, instead of just 1440/1080/720/etc, although the basic selector still includes the generic sizes. At least it's good to know that Opus will use a wide quality range; that's good news for modules and chiptunes that need quite a bit more.

Nintendo Maniac 64

23rd October 2015, 01:53

Oh god, I can't stop listening.

Addictive song? :p

That video brings up some very strange frame-sizes on the downloader

The utility I use, CompleteYouTubeSaver, always gave the exact video resolution for the DASH streams (except 144p) and not just a generic maximum resolution like it does for non-DASH streams. Besides, the vertical resolution isn't that strange - the 224 in 640x224 is in fact the same output resolution of most NES and SNES games, and is still in fact dividable by 16; the only really strange ones are those with a width of 854px and 426px but that's understandable since they were always kind of a weird case of trying to get 16:9 resolutions that matched the non-widescreen resolutions of 480p and 240p (which aren't even dividable by 9).

One could also include 256px, but that's simply a case of having a "root value" (or whatever it's called) of 2 (2^5 for 128) rather than a root value of 10 (10 x 2^4 for 160), and if you know anything about resolutions and how well they match up to other resolutions, you'd know that 128 and 160 aren't a multiple of each other until you reach 640, therefore anything that has a width smaller than that can give weird fractional resolutions.

Anyway, something else I discovered - it seems YouTube's AAC encoder on that video is using a very recent version of ffmpeg (read: not even a month old).

Therefore, I believe it includes the considerable quality improvements that were mentioned during the VideoLAN Dev Days AAC presentation: https://www.youtube.com/watch?v=QiV60eGY11o

In particular, I find that the 192kbps AAC sounds noticably better than the Opus encode even though the source audio was even uploaded at 32KHz which would resample much more nicely into the 48KHz that Opus uses rather than the 44.1KHz used by AAC.

foxyshadis

23rd October 2015, 02:13

YouTube Video and Audio Downloader shows 1344p, 896p, 672p, 448p, 298p, 224p, 150p, and 90p. It's... quite specific. Previously I believe it only showed the closest "standard" frame size, as of a year ago, anyway.

Nintendo Maniac 64

23rd October 2015, 02:37

NikosD

23rd October 2015, 07:50

Using latest Chrome v47.0.2526.27 beta-m (64-bit) and my Core i7-4790 on FTTC+VDSL2 50Mbps line, it is impossible to play in real-time Youtube 4K60fps VP9 clips.

I tried both Window and Full Screen mode and I had stuttering, crippled sound and dropped frames.
The same thing happened with 4K50fps clips, but 4K30fps and 1440p60 fps had <50% CPU utilization and played fine.

I thought that I have to blame the VP9 decoding engine of Chrome 64bit because I supposed the same 4K60fps clip from Youtube could easily be decoded by MPC-HC x64 v1.7.9 off-line.

Guess what...It can't.

Using latest nightly x64 v1.7.9.190 and EVR-CP renderer, I have almost the same CPU utilisation and dropped frames on a 4K60fps 18Mbps VP9 clip, with crippled sound and no realtime decoding capability (I get around ~50fps)

I managed to play that file in realtime with no dropped frames and normal sound only using EVR renderer.

It is possible that with a discrete GPU could play that file in EVR-CP, but I'm not sure if I don't try it.

My full specs:
Win 10 x64 - MPC-HC x64 v1.7.9.190 - Core i7 4790 - iGPU HD 4600

Clip:
3840 x 2160@60fps VP9 at 18Mbps

I think we need HW acceleration for those clips.

nevcairiel

23rd October 2015, 08:48

If changing the video renderer allows them to play, it sounds like the processing/scaling of the video caused the problems, not the decoding alone, so having hardware acceleration wouldn't necessarily help. Default EVR uses the built-in Hardware Scaler in the GPU - EVR-CP cannot use that and uses Pixel Shaders instead, which may have overwhelmed the GPU.
What Chrome uses to process and scale the video I couldn't tell you.

Speaking of Hardware Acceleration - unfortunately Haswell systems won't get VP9 acceleration from the looks of it, only Broadwell and up.

NikosD

23rd October 2015, 09:22

The behavior of the whole Intel based system changes, when you put a discrete card and performance rises significantly as tests show, mainly because of system memory been used by iGPU.

That kind of performance increase is separated from the discrete GPU performance alone, which is added to the equation.

The EVR - CP performance drops for all Intel iGPU I have tested so far (from Sandy to Haswell) and it's not so easily explained by GPU load of the pixel shaders.

Because GPU load is always very, very low on Haswell HD4600 even when scaling 4K to a small window.

Also 4K30 fps has no problem at all.

I haven't managed to explain it yet, so I take it as just a fact.

All I know is that it comes critical only for clips pushing the decoding performance to the edge.

Regarding HW acceleration, judging by similar cases for H.264 like 4K120fps, it might couldn't help.

Because for 4K@120fps H.264 I have exactly the same problem with EVR CP vs EVR like 4K60 fps VP9.

It seems that frame rate and not resolution causes issues with iGPU and EVR-CP with or without HW acceleration.

But H.264 4K60 fps is a piece of cake for Haswell.

Regarding HW acceleration of VP9 for Haswell, I have already told you that is not in Intel's intentions to do it - in hybrid mode of course.

Because I'm not sure if there is any chip of PC used GPUs, that has native HW VP9 support inside - not hybrid which is driver based.

I wrote about HW acceleration of VP9 in general, because I see that VP9's CPU decoders don't look optimised even for 64bit.

MoSal

23rd October 2015, 13:17

Anyway, something else I discovered - it seems YouTube's AAC encoder on that video is using a very recent version of ffmpeg (read: not even a month old).

Therefore, I believe it includes the considerable quality improvements that were mentioned during the VideoLAN Dev Days AAC presentation: https://www.youtube.com/watch?v=QiV60eGY11o

In particular, I find that the 192kbps AAC sounds noticably better than the Opus encode even though the source audio was even uploaded at 32KHz which would resample much more nicely into the 48KHz that Opus uses rather than the 44.1KHz used by AAC.

Oh boy.

1- YouTube never used lavc's still experimental AAC encoder.

2- Even with all the improvements. lavc's encoder is still not a very high-quality one.
Apple's encoder is considered the best. Fraunhofer's offerings (both open and closed source) are competitive. No other encoder comes close.

3- Opus has one (known) encoder*. And it's considered better than Apple's AAC encoder. Both are considered transparent for most people with the majority of samples at the bitrates > 128kbps. The quality of MP3s from the 90s is long gone!

4- If you use a blind testing tool. You would realize how powerful the placebo effect can be.

5- Resampling from 32KHz to either 48KHz or 44.1KHz should be transparent with any half-decent resampler.

* Note: libopus provides speed profiles (complexity from 0 to 10). 10 (high-complexity) is the default. It's fast. And I don't think anyone ever felt the need to lower that complexity.

Nintendo Maniac 64

24th October 2015, 01:12

Using latest nightly x64 v1.7.9.190 and EVR-CP renderer, I have almost the same CPU utilisation and dropped frames on a 4K60fps 18Mbps VP9 clip, with crippled sound and no realtime decoding capability (I get around ~50fps)

Enable "D3D Fullscreen" - that will give you the lowest CPU utilization possible and is light enough on the GPU that it is in fact the only way to get 1080p 60fps to run smoothly on nearly 10-year old Intel integrated GPUs (like the Intel 965GMA).

Also, you may want to check the Windows task manager and see if the VP9 decoder used by MPC-HC is not running at less than 50% CPU utilization, because if it is then it means the VP9 decoder simply isn't taking full advantage of your 4 CPU cores and that could explain the performance (in particular I get ~42fps with 3840x2160 60fps VP9 videos on "only" my dual core Pentium G3258 @ 4.6GHz with ~100% CPU utilization).

If you want, you could try to update the LAVfilters in MPC-HC 64bit to v0.66; first make sure MPC-HC is closed, navigate to your MPC-HC 64bit program folder, then go into the "LAVFilters64" folder and delete everything in it - don't close the folder window, you'll need it again. Now download the LAVFilters-0.66-x64.zip from the link below and extract all the files from the ZIP into the "LAVFilters64" folder that you left open - make sure you extract the files directly and not into a sub folder. That's it, no configuration or anything needed.

https://github.com/Nevcairiel/LAVFilters/releases/tag/0.66

----------------------------------------------------------------

4- If you use a blind testing tool. You would realize how powerful the placebo effect can be.

I did do a blind test - it's definitely not placebo because until I turned my monitor back on, I thought the track that sounded higher quality even with lower volume was for sure going to be the Opus track, but when I turned my monitor back it was revealed that the track that I thought sounded better was actually the AAC track, not the Opus one.

For reference, my configuration is to use foobar2000 + ASIO (so no resampling on my Xonar DS), both tracks have replaygain metadata, playback order set to "Repeat (track)"; from there I put both audio tracks into a single playlist in foobar2000, start playing one of the audio tracks, turn off my monitor, then press the "Page Down" key (keyboard shortcut for "next track") a bajillion times until I've completely lost count of what track is playing at a given time and then I just press "Page Down" again to switch between tracks, once I've chosen which one sounds better I let it play and then turn the monitor back on and see which track is actually playing.

After that blind test I compared the AAC and Opus tracks to the original FLAC via Audacity by lining up the waveforms, doing an invert, and mixing the waveforms together, and the 192kbps AAC track was definitely more similar to the original FLAC than the Opus version was to the same FLAC; in other words, the AAC track when inverted and mixed with the FLAC gave a considerably quieter resulting waveform than when I mixed the inverted Opus track with the FLAC.

wanezhiling

24th October 2015, 03:23

Using latest nightly x64 v1.7.9.190 and EVR-CP renderer, I have almost the same CPU utilisation and dropped frames on a 4K60fps 18Mbps VP9 clip, with crippled sound and no realtime decoding capability (I get around ~50fps)
What's your memory? single or dual channel? That's maybe the point. On my laptop, the channel has an enormous impact on video playback...

example
single: 99% cpu load, 3840x2160_h264_8-bit_40m_30fps, sw decoding, evr-cp
dual: 15% cpu load, 3840x2160_h264_8-bit_40m_30fps, sw decoding, evr-cp

NikosD

24th October 2015, 07:32

Enable "D3D Fullscreen" - that will give you the lowest CPU utilization possible and is light enough on the GPU that it is in fact the only way to get 1080p 60fps to run smoothly on nearly 10-year old Intel integrated GPUs (like the Intel 965GMA).

It does make a little difference, but still dropped frames and lower than 60fps.

Also, you may want to check the Windows task manager and see if the VP9 decoder used by MPC-HC is not running at less than 50% CPU utilization, because if it is then it means the VP9 decoder simply isn't taking full advantage of your 4 CPU cores and that could explain the performance

I have a Windows gadget always open, monitoring CPU, GPU, HDD, SSD etc.
The CPU utilization is about 40% - 60% for all 8 threads.

If you want, you could try to update the LAVfilters in MPC-HC 64bit to v0.66; first make sure MPC-HC is closed, navigate to your MPC-HC 64bit program folder, then go into the "LAVFilters64" folder and delete everything in it - don't close the folder window, you'll need it again. Now download the LAVFilters-0.66-x64.zip from the link below and extract all the files from the ZIP into the "LAVFilters64" folder that you left open - make sure you extract the files directly and not into a sub folder. That's it, no configuration or anything needed.

https://github.com/Nevcairiel/LAVFilters/releases/tag/0.66

MPC-HC v64 1.7.9.190 has LAV filters 0.66 built-in.

Also, I always grab the latest official nightly build from here:
http://files.1f0.de/lavf/nightly/

Latest nightly is 0.66.31 right now.

What's your memory? single or dual channel? That's maybe the point. On my laptop, the channel has an enormous impact on video playback...

example
single: 99% cpu load, 3840x2160_h264_8-bit_40m_30fps, sw decoding, evr-cp
dual: 15% cpu load, 3840x2160_h264_8-bit_40m_30fps, sw decoding, evr-cp

Dual channel of course :)

(in particular I get ~42fps with 3840x2160 60fps VP9 videos on "only" my dual core Pentium G3258 @ 4.6GHz with ~100% CPU utilization).

It's time for a small test.

For anyone interested, I downloaded this 4K60fps VP9 clip from YouTube.
The first 70s is here:
https://www.sendspace.com/file/o0oe5t

Using my system above Win 10 x64 - MPC-HC x64 v1.7.9.190 - Core i7 4790 - iGPU HD 4600, I always get a few (1 to 4) dropped frames in the first 70s, the CPU utilization goes up to 70% (8 threads), the sound silences a few times and the frame rate drops to ~52fps a few times.

What are your results using EVR-CP?

MoSal

24th October 2015, 08:29

I did do a blind test - it's definitely not placebo because until I turned my monitor back on, I thought the track that sounded higher quality even with lower volume was for sure going to be the Opus track, but when I turned my monitor back it was revealed that the track that I thought sounded better was actually the AAC track, not the Opus one.

For reference, my configuration is to use foobar2000 + ASIO (so no resampling on my Xonar DS), both tracks have replaygain metadata, playback order set to "Repeat (track)"; from there I put both audio tracks into a single playlist in foobar2000, start playing one of the audio tracks, turn off my monitor, then press the "Page Down" key (keyboard shortcut for "next track") a bajillion times until I've completely lost count of what track is playing at a given time and then I just press "Page Down" again to switch between tracks, once I've chosen which one sounds better I let it play and then turn the monitor back on and see which track is actually playing.

That'a not how you do (double-)blind tests. For starters, the levels must be matched.

You can use this component in foobar2000:
http://www.foobar2000.org/components/view/foo_abx

After that blind test I compared the AAC and Opus tracks to the original FLAC via Audacity by lining up the waveforms, doing an invert, and mixing the waveforms together, and the 192kbps AAC track was definitely more similar to the original FLAC than the Opus version was to the same FLAC; in other words, the AAC track when inverted and mixed with the FLAC gave a considerably quieter resulting waveform than when I mixed the inverted Opus track with the FLAC.

That's not a valid way to evaluate lossy codecs. You wouldn't use PSNR to evaluate
video quality, would you?

wanezhiling

24th October 2015, 09:47

For anyone interested, I downloaded this 4K60fps VP9 clip from YouTube.
The first 70s is here:
https://www.sendspace.com/file/o0oe5t

Using my system above Win 10 x64 - MPC-HC x64 v1.7.9.190 - Core i7 4790 - iGPU HD 4600, I always get a few (1 to 4) dropped frames in the first 70s, the CPU utilization goes up to 70% (8 threads), the sound silences a few times and the frame rate drops to ~52fps a few times.

What are your results using EVR-CP?
Similarly here, can't watch it fluently at all. http://i.imgur.com/Aymeu6T.png

But PotPlayer x64 seems just fine, stable 60 all the time. http://i.imgur.com/HaQNOlT.png

i7 4710MQ@2.5G
HD 4600 with latest 4294 driver
16g(8*2) Dual channel
win 8.1 with up3
MPC-HC x64 1.7.9.190
PotPlayer x64 1.6.56209

NikosD

24th October 2015, 09:53

Similarly here, can't watch it fluently at all. http://i.imgur.com/Aymeu6T.png

But PotPlayer x64 seems just fine, stable 60 all the time. http://i.imgur.com/HaQNOlT.png

Can you enable everything from the View menu of MPC-HC like statistics, information etc and watch dropped frames and framerate.

My experience is very smooth, but I see occasionally dropped frames and framerate, the sound silences at moments and the CPU utilization never goes more than 70%.

The VP9 decoder probably is not optimized for 8 threads.

wanezhiling

24th October 2015, 09:58

Can you enable everything from the View menu of MPC-HC like statistics, information etc and watch dropped frames and framerate.
terrible. http://i.imgur.com/cRjsY4t.png

wanezhiling

24th October 2015, 11:12

Similarly here, can't watch it fluently at all. http://i.imgur.com/Aymeu6T.png

But PotPlayer x64 seems just fine, stable 60 all the time. http://i.imgur.com/HaQNOlT.png

Found the reason, PotPlayer outputs YV12 by default on iGPU and MPC-HC(LAV) outputs NV12 by default on all GPUs.

@NikosD, try to disable LAV's NV12(then will auto use YV12) and you will get something surprised:p

NikosD

24th October 2015, 17:36

@NikosD, try to disable LAV's NV12(then will auto use YV12) and you will get something surprised:p

I really, really can't believe in my own eyes!

This is something unbelievable, it's the first time that I can play everything in EVR-CP without problem using my iGPU.

By disabling NV12, LAV video uses YUY2 and the 4K60fps VP9 clip above can be played at around 20-30% CPU usage using only 2 to 4 cores, depending on the bitrate.

But when seeking, I can see just for a sec, the CPU usage going even 70% using all 8 threads.

During normal playback, it's always a steady 60fps realtime playback with no problem in sound at all.

I can even decode those 4K120fps H.264 using EVR-CP I was complaining in my previous post.

You solved an issue I had for the last 2 years!

I owe you a favor, at least!

I don't know if LAV Video can or should do something about it, like disabling NV12 for iGPU.

Or Intel.

nevcairiel

24th October 2015, 17:49

NV12 is the correct format to use for perfect unmolested quality, and its in general the most GPU friendly 8-bit 4:2:0 format. Its also the same format that a hardware accelerator decodes to.
My guess is that disabling it forces the GPU (or something inside EVR-CP) to do something different, like using a lower quality scaling algorithm, which boosts performance at the expense of quality, or something like that.

YUY2 is a 4:2:2 format, which means LAV has to upsample chroma, which is never good for quality.
Perfect bit-exact quality is always LAVs primary goal, and NV12 ensures that.

If it only affects EVR-CP, and normal EVR works fine, it might as well be a problem in MPC-HCs renderer code.
Might make sense to start poking there.

wanezhiling

24th October 2015, 19:07

By disabling NV12, LAV video uses YUY2
Hmmm.. YV12 should be automatically used when NV12 is disabled only, at least here.

@nevcairiel, YV12 won't lose quality in such situation, right?:)

Nintendo Maniac 64

25th October 2015, 04:27

@NikosD, try to disable LAV's NV12(then will auto use YV12) and you will get something surprised:p

O_o What the crap, this let's me play 3840x2160 VP9 videos at 56fps - that's a 33% increase over my previous 42fps! I'm a beta tester for SVP, and any and all extra CPU headroom on video playback is very important for such a program, so I'm definitely going to have to do more testing regarding this with various different hardware and possibly report this back to the SVP devs...

For reference I'm also using Haswell integrated graphics.

@nevcairiel, YV12 won't lose quality in such situation, right?:)

I would definitely like to know this as well since SVP currently can only handle 8bit 4:2:0 video signals as an input.

----------------------------------------------------------------

MoSal, just want to let you know that, several months ago (maybe even a year?) an audio-focused discussion got removed from one of the VP9 threads on Doom9 because the mods deemed it as off-topic. I had a lengthy post typed up and even posted, but I've since edited it out due to the fear of off-topic-ness.

I will at least say the following quick things:

1. That utility it does not work well with ASIO without a hardware volume control and also doesn't account for AAC's extra bit of silence at the beginning.

2. Even with a successful ABX test it only tells you that I can tell the difference, not which is higher quality than the other.

3. I said I applied ReplayGain to both tracks via foobar2000.

4. I mentioned volume because humans perceive louder volumes as sounding better.

5. Of course the inverted waveform comparison isn't really valid, that's why I did it after my testing.

NikosD

25th October 2015, 06:52

My guess is that disabling it forces the GPU (or something inside EVR-CP) to do something different, like using a lower quality scaling algorithm, which boosts performance at the expense of quality, or something like that.

It is true that when the camera zooms in from 28 sec to 39 sec, the picture in windows mode looks like a little weird around the grass, like transparent water floating around.

In full screen (1920x1080), everything looks good.

If it only affects EVR-CP, and normal EVR works fine, it might as well be a problem in MPC-HCs renderer code.
Might make sense to start poking there.

Using EVR and disabling NV12, nothing really changes.
There are no artifacts around the grass, like in EVR-CP and no performance increase.

I can't see what is the output type, because EVR doesn't allow Ctrl-J option of MPC-HC which needs a layer over the image.

Is there any other way to see what output type MPC-HC uses with EVR renderer ?

MoSal

25th October 2015, 09:06

MoSal, just want to let you know that, several months ago (maybe even a year?) an audio-focused discussion got removed from one of the VP9 threads on Doom9 because the mods deemed it as off-topic. I had a lengthy post typed up and even posted, but I've since edited it out due to the fear of off-topic-ness.

Okay.
We can stop here, or move the discussion to HydrogenAudio if you like.

I will at least say the following quick things:

1. That utility it does not work well with ASIO without a hardware volume control and also doesn't account for AAC's extra bit of silence at the beginning.

Maybe.
I don't personally use foobar2000, or Windows for that matter.

2. Even with a successful ABX test it only tells you that I can tell the difference, not which is higher quality than the other.

The mere presence of an audible difference would be surprising to me at those bitrates.

3. I said I applied ReplayGain to both tracks via foobar2000.

But why did you need to do this. They both come from the same source and their loudness is very close (1.74 dB vs. 1.72 dB using loudness-scanner). There could be something wrong with foobar2000 or the way you use it!

4. I mentioned volume because humans perceive louder volumes as sounding better.

Not in the presence of clipping. And if your Opus stream plays much louder than the AAC stream, then there is a huge gain applied that could be causing clipping.

wanezhiling

25th October 2015, 09:36

I can't see what is the output type, because EVR doesn't allow Ctrl-J option of MPC-HC which needs a layer over the image.

Is there any other way to see what output type MPC-HC uses with EVR renderer ?
You did a totally wrong way to see the output type, and now I knew why you said 'YUY2' :p
Correct way: step1 (http://i.imgur.com/8PuWDEy.png), step2 (http://i.imgur.com/PyrL2d1.png)
If it only affects EVR-CP, and normal EVR works fine, it might as well be a problem in MPC-HCs renderer code.
No.
It doesn't matter which renderer is used.
'NV12+vanilla EVR' is much better than 'NV12+EVR-CP', that's true.
But 'NV12+vanilla EVR' is still much heavier than 'YV12+vanilla EVR', and 'NV12+vanilla EVR' is even a litte heavier than 'YV12+EVR-CP'!!
The 'NV12' colorspace is just horrible for iGPUs.....:o

==========
With Nintendo Maniac 64's another verification, now we can make a conclusion: all media player should use YV12 by default instead of NV12 on Intel platform in order to avoid significant performance loss.

Aleksoid1978

25th October 2015, 10:27

The 'NV12' colorspace is just horrible for iGPUs.....:o

Maybe you try to write about this "trouble" to Intel developers/engineers :)

NikosD

25th October 2015, 11:40

You did a totally wrong way to see the output type, and now I knew why you said 'YUY2' :p
Correct way: step1 (http://i.imgur.com/8PuWDEy.png), step2 (http://i.imgur.com/PyrL2d1.png)

No.
It doesn't matter which renderer is used.
'NV12+vanilla EVR' is much better than 'NV12+EVR-CP', that's true.
But 'NV12+vanilla EVR' is still much heavier than 'YV12+vanilla EVR', and 'NV12+vanilla EVR' is even a litte heavier than 'YV12+EVR-CP'!!
The 'NV12' colorspace is just horrible for iGPUs.....:o

You are right, thanks!

And now it's the time to do my magic.

I tested all output video formats (NV12, YV12, YUY2) using DXVA Checker v3.9.0 with LAV filters 0.66.31 in Decode (rendereless) and Playback (vanilla EVR) modes.

Win x64 - Core i7-4790 - iGPU HD4600@1.5GHz - v.4279

Test clip:
Big Buck Bunny -70s (4K60fps VP9)

(I have never thought before this test, that I had to benchmark output modes!)

GPU usage is lowest all the time (only 600MHz lowest clock and GPU usage <20%)

Decode - Renderless

64bit

NV12/YV12 83/112/175 CPU usage 72%
YUY2 84/110/163 CPU usage 71%

32bit

NV12/YV12 73/106/191 CPU usage 73%
YUY2 74/104/160 CPU usage 72%

Playback - Vanilla EVR (scale to 1280 x 720)

64bit

YV12 79/100/139 CPU usage 73%
NV12 65/74/86 CPU usage 52%
YUY2 43/50/56 CPU usage 35%

32bit

YV12 66/92/140 CPU usage 75%
NV12 61/72/87 CPU usage 61%
YUY2 45/51/56 CPU usage 41%

Comments:

1) CPU VP9 x64 and x86 decoders are close to performance for FFMpeg (LAV Video), it's not like HEVC where x64 is a lot faster.
That could mean lack of optimizations for VP9 x64.

2) In rendereless mode (Decode) all output formats are very close and YV12 and NV12 have exactly the same performance.
(Actually NV12 is 1 frame faster than YV12 but I didn't want to write another line just for that)

3) Using vanilla EVR (Playback) there is a big performance drop for NV12 and a huge one for YUY2.
That drop of performance is followed by a proportional CPU usage drop.

4) I think EVR-CP could have the same results, but with even lower values or it could be a lot different - I don't know and I don't know a way on how to test it.

5) Until Intel releases a NV12 more efficient driver, I think LAV Video could disable NV12 for iGPUs (if it's possible)

wanezhiling

25th October 2015, 11:59

And not for VP9, but for other common formats like h264, there is still a hugh difference between NV12 and YV12 on Intel platform.
I remember that before I upgraded to a 16gb dual channel memory, I used a single 8gb, when playing some common 4k h264 30p clips under evr-cp (sw decoding), YV12 had a normal result (low cpu usage and stable fps), while nv12 is terrible too(mpc-be's devs may remember that.)

nevcairiel

25th October 2015, 15:11

There is not enough information to act on anything. NikosD himself said that disabling NV12 changed the image for the worse in one scene. If that is indeed the case then my earlier attempt to explain it may indeed be correct and the iGPU uses a faster and lower quality scaler with YV12 input, potentially because the high-quality one needs NV12.
From what I remember from years ago, proper hardware deinterlacing also required NV12 input on Intel GPUs .. but thats been a few generations, so who knows if thats still true today.

NikosD

25th October 2015, 15:15

I will take that table i made above and put it in a new thread in Intel forums and see what happens.

Hopefully in a year or two, Intel will fix it.

wanezhiling

25th October 2015, 15:31

How to explain dxva situation? dxva also means NV12, but no such performance loss

NikosD

25th October 2015, 16:01

There is performance loss.

As I wrote above, using YV12 it's the first time I can play some difficult 4K120fps H.264 clips.

I opened the thread in Intel's forum.

De-M-oN

26th October 2015, 01:42

I'm a beta tester for SVP

Can you PLEASE tell them, that "resampling" shouldnt be enabled as default !!! ?

It results in heavy ghosting if some user again forgot to set correct fps and so on in project settings and ghosting looks awful. On a digicam as well. Is there any reason why you would torture your video with ghosting? And any reason for this being default?

I cant stand anymore these thousands of videos uploaded to youtube with ghosting (and maybe even said in project settings the video being interlaced) while you directly know: This user has coded with sony ..

Also: Why does the software not reading out the details itself as default? Why so intrusive with presets, especially for an input file..

Nintendo Maniac 64

26th October 2015, 02:36

NikosD himself said that disabling NV12 changed the image for the worse in one scene. If that is indeed the case then my earlier attempt to explain it may indeed be correct and the iGPU uses a faster and lower quality scaler with YV12 input

What if I'm not doing any upscaling? (that is, MPC-HC has the video image set to "Normal size")

----------------------------------------------------------------

Can you PLEASE tell them, that "resampling" shouldnt be enabled as default !!! ?

I am not sure what you mean by "resampling"... Could you please elaborate?

De-M-oN

26th October 2015, 03:32

Vegas calls his fps conversion method via blending the frames instead of deleting/inserting them "resampling" and this is enabled for default. Thats why you see so many ghosting videos on youtube.

vivan

26th October 2015, 04:39

SVP is not "Sony Vegas Pro", it's SmoothVideo Project (https://www.svp-team.com).

And, yeah, I agree about sony vegas.

Nintendo Maniac 64

26th October 2015, 05:06

UPDATE: SVP already uses YV12 by default, so the performance issue in question does not apply to it.

I did notice however that, on my old Intel 965GMA integrated graphics, the performance of NV12 matches the performance of YV12.

----------------------------------------------------------------

SVP is not "Sony Vegas Pro", it's SmoothVideo Project (https://www.svp-team.com)

Ohhhhhhhhhhhhhh.

Yeah, uh, I've never used Sony Vegas. :p

Interestingly enough, resampling the framerate by doing blending rather than repeating/deleting frames is actually greatly preferred if you're a user of SVP because the extra smoothness ends up just amplifying the difference between repeated/deleted frames and normal frames thereby resulting in quite the annoying judder. However, I must say that the judder is considerably less noticeable when doing 30--to->24fps compared to the likes of 25--to->24fps or 24/25--to->30fps.

AVC, VP8, HEVC, and VP9 all support variable framerates, so why not take advantage of that function rather than re-encoding everything a specific static framerate? Heck new monitors can even handle variable refresh rates in real time!

Nintendo Maniac 64

6th November 2015, 17:07

I'm visiting in Penn State again, so you know what that means - uploading more lossless test video footage to YouTube for quality testing.

This time it's the same video as before (lossless 1080p) but at 2x speed for a cheaty 60fps (because at the time of recording my PC wasn't fast enough to capture true 60fps).

Link: https://www.youtube.com/watch?v=NXPADDVmXU4

UPDATE: I've also re-uploaded a copy of the normal speed 1080p 30fps version of that video:
https://www.youtube.com/watch?v=B7Y186Wh_AE

I'll analyze the quality differences and stuff at a later date.

mzso

24th December 2015, 21:17

LithosZA

1st January 2016, 09:05

Did the 1920*1200 resolution disappear. Not long ago youtube played 1920*1200 when available but showed 1080p as a resolution. Now I don't get it at all.
You can still download it with downloader tools though. So they weren't actually removed, yet.

There isn't any non-DASH 1080p streams available anymore.
You can go to https://www.youtube.com/html5 to see if your browser will be able to playback 1080p.
You need 'MSE & H.264' to see H.264 1080p streams
and
You need 'MSE & WebM VP9' to see VP9 1080p streams.

The DASH video and audio streams are separated while the old non-DASH streams are combined.

mzso

1st January 2016, 13:02

There isn't any non-DASH 1080p streams available anymore.
You can go to https://www.youtube.com/html5 to see if your browser will be able to playback 1080p.
You need 'MSE & H.264' to see H.264 1080p streams
and
You need 'MSE & WebM VP9' to see VP9 1080p streams.

The DASH video and audio streams are separated while the old non-DASH streams are combined.

That's not the issue, also I see 1080p. (Actually it's not true, some non-dash streams are still available)

Actually it turns out it's the fault of an addon of mine. Which somehow prevents the switch to 1200p

De-M-oN

1st January 2016, 13:03

Maybe it forces to use H.264 and/or flash Player.

Because almost the whole mankind still didnt understood that their VP9 encodes look better than H.264.

mzso

1st January 2016, 13:17

Maybe it forces to use H.264 and/or flash Player.

Because almost the whole mankind still didnt understood that their VP9 encodes look better than H.264.

No, I have vp9 disabled in the browser. But it's not the problem.
Also VP9 is not better with the bitrates they use. Usually comparable or worse. But it's most certainly much sh*ttier to play back with the lack of HW acceleration and decoder optimization.

De-M-oN

1st January 2016, 13:19

No, I have vp9 disabled in the browser. But it's not the problem.

Thats the problem. HFR 1440 and HFR 4k are only offered as VP9. This includes 1920x1200.

VP9 is very visible better in any case. Also at 1080p very much better already.

edit: Oh and dont use Firefox. It's VP9 Decoder is the slowest the world have seen. No wonder it is default off.
Chrome can handle 4k60 perfectly fine for me, while firefox already framedrops at 1080p60..

mzso

1st January 2016, 16:40

Thats the problem. HFR 1440 and HFR 4k are only offered as VP9. This includes 1920x1200.

VP9 is very visible better in any case. Also at 1080p very much better already.

edit: Oh and dont use Firefox. It's VP9 Decoder is the slowest the world have seen. No wonder it is default off.
Chrome can handle 4k60 perfectly fine for me, while firefox already framedrops at 1080p60..

Wrong. You're information might be outdated. The same resolutions are available with AVC.

Also for a long while now 1920*1200 is tied into 1080p, and the change occurs automatically.

VP9 is very visible better in any case. Also at 1080p very much better already.

edit: Oh and dont use Firefox. It's VP9 Decoder is the slowest the world have seen. No wonder it is default off.
Chrome can handle 4k60 perfectly fine for me, while firefox already framedrops at 1080p60..

I see you're a vp9 "enthusiast". Anyway it was never the case whenever I checked. I only saw some edge cases that showed significant quality benefit for VP9. Probably shared in this very topic.

Also My PC is not the newest so even in chrome I tend to not get adequate performance with vp9.

De-M-oN

1st January 2016, 16:48

Wrong. You're information might be outdated. The same resolutions are available with AVC.

No they arent.

Do this URL

https://www.youtube.com/watch?v=l-FVjqlNUiY

into this:

http://www.h3xed.com/blogmedia/youtube-info.php

And you'll see that 1920x1200 is only available as VP9 and runs under the Label 1080p but runs Encode 308 there. 303 is still encoded and played back by automatic, but if you choose manually the 1080p Label it accesses the 308 encode. 308 only runs as seperate label if you upload native 1440p.

1440 and 4k are only as H.264 available if you dont use higher than 40 fps. And the H.264 variants look worse.

I see you're a vp9 "enthusiast"
I just use that, what looks better. And H.264 looks much worse.

Anyway it was never the case whenever I checked
That is highly unbelievable. But ok.

edit: Here the table which I did:

http://www.letsplayforum.de/index.php/Thread/159108-Supernoob-Frage-Bandicam/?postID=2163984#post2163984

That should make things more clear about labels and encodes.

foxyshadis

1st January 2016, 17:05

No they arent.

Do this URL

https://www.youtube.com/watch?v=l-FVjqlNUiY

into this:

http://www.h3xed.com/blogmedia/youtube-info.php

And you'll see that 1920x1200 is only available as VP9 and runs under the Label 1080p but runs Encode 308 there. 303 is still encoded and played back by automatic, but if you choose manually the 1080p Label it accesses the 308 encode. 308 only runs as seperate label if you upload native 1440p.

1440 and 4k are only as H.264 available if you dont use higher than 40 fps. And the H.264 variants look worse.

Your site must be outdated, I see the H.264 1200p is advertised and available as format 264 (https://r3---sn-a5meknel.googlevideo.com/videoplayback?expire=1451685189&keepalive=yes&requiressl=yes&nh=IgpwcjAxLmxheDE2KgkxMjcuMC4wLjE&mt=1451663458&ipbits=0&sparams=clen,dur,gir,id,initcwndbps,ip,ipbits,itag,keepalive,lmt,mime,mm,mn,ms,mv,nh,pl,requiressl,source,upn,expire&mime=video/mp4&upn=THGDIpJkX74&signature=8F83B2D09F24D1141900B408A96772A5305F277D.2CCBDFAA8367D1B322B0FDA26133FD4941AE3600&initcwndbps=1827500&itag=264&sver=3&clen=1263180108&gir=yes&pl=33&fexp=9407537,9416126,9419891,9420452,9420736,9422596,9423237,9423662,9423715,9423974,9425490,9425782,9426012,9426054,9426516&lmt=1451539312084361&ip=2601:201:8201:1e03:2969:aae7:435d:836d&key=yt6&ms=au&mv=m&source=youtube&id=o-AAFMfIKcgoN1wiPjdmP897SqdzAc8EU5rDmP1ViWQawI&dur=1424.166&mn=sn-a5meknel&mm=31&ratebypass=yes&title=Let%27s%20Play%20-%20DooM%202%3A%20Valiant%20%40%20UV%20--%20MAP31%20%7B%20Cyberwar%207734%20%7D%20%5B2%2Fx%5D). Quality is a bit meh, but it's leaps and bounds better than 1080p. Of course, as decent as the 6.5mbps H.264 and VP9 are, they pale compared to the 10mbps VP9 format 308.

De-M-oN

1st January 2016, 17:11

This has just 30fps.

Read what I wrote: I said 1440 and 4k HFR(!) are only available as VP9.

The h3xed.com thingy accesses directly youtube, so if there is not more than that listed - there is then no more.

edit: For quality comparison you better use this (makes it easier to see differences):

https://www.youtube.com/watch?v=mbsR82lJKMs

Because the Doom Video is quite good compressable.

mzso

1st January 2016, 19:50

No they arent.

Do this URL

https://www.youtube.com/watch?v=l-FVjqlNUiY

into this:

http://www.h3xed.com/blogmedia/youtube-info.php

And you'll see that 1920x1200 is only available as VP9 and runs under the Label 1080p but runs Encode 308 there. 303 is still encoded and played back by automatic, but if you choose manually the 1080p Label it accesses the 308 encode. 308 only runs as seperate label if you upload native 1440p.

I just use that, what looks better. And H.264 looks much worse.

That is highly unbelievable. But ok.

edit: Here the table which I did:

http://www.letsplayforum.de/index.php/Thread/159108-Supernoob-Frage-Bandicam/?postID=2163984#post2163984

That should make things more clear about labels and encodes.

Looks like it's the standard case of youtube works differently for different videos (and people)

1200p works for the video I tried with: https://www.youtube.com/watch?v=n5qJ1Daf5R4

Also 1200p is downloadable for yours too. Even though the yt player does not play it.

1440 and 4k are only as H.264 available if you dont use higher than 40 fps.
This Might be true, but you weren't talking about this until now...

And the H.264 variants look worse.

I just use that, what looks better.

And H.264 looks much worse.

That is highly unbelievable. But ok.

Don't be such a zealot.

Here's a screenshot of a random video (https://www.youtube.com/watch?v=eQ10N8TRZvk)(middle of it): http://screenshotcomparison.com/comparison/156397
With the ugly-ass artifacts I'd argue that it's a lot worse.
It's even worse in motion.

mzso

1st January 2016, 19:56

By the way I don't get VP9 in firefox even though I enabled all the webm options in about:config and disabled vp9. Does anyone know what's wrong?
https://www.youtube.com/html5 also shows it to be available.

De-M-oN

1st January 2016, 20:37

Looks like it's the standard case of youtube works differently for different videos (and people)
No. Only one thing: fresh new channels often havent directly VP9 available and get only H.264.

1200p works for the video I tried with: https://www.youtube.com/watch?v=n5qJ1Daf5R4

Also 1200p is downloadable for yours too. Even though the yt player does not play it.
It does. The label is just 1080. But it plays 308. I told you already in previous post and even linked you a table.

This Might be true, but you weren't talking about this until now...
Oh yes I did several times now.

One of the examples where I did:

Thats the problem. HFR 1440 and HFR 4k are only offered as VP9. This includes 1920x1200.

Don't be such a zealot.

Here's a screenshot of a random video (https://www.youtube.com/watch?v=eQ10N8TRZvk)(middle of it): http://screenshotcomparison.com/comparison/156397
With the ugly-ass artifacts I'd argue that it's a lot worse.
It's even worse in motion.

And here is my:

http://screenshotcomparison.com/comparison/156406

My assumption is, that you play 1920x1200 @ 30 fps AVC and compare it with 1920x1080 @ 60 VP9, just because you think it is the same, because of same LABEL at the player.

But look on my table I gave you in my previous post.

This is comparison between 1920x1080 @ 60 H.264 AVC vs 1920x1080 @ 60 VP9.

The used video is this one:

https://www.youtube.com/watch?v=mbsR82lJKMs

mzso

1st January 2016, 22:15

It does. The label is just 1080. But it plays 308. I told you already in previous post and even linked you a table.

It plays 1200p AVC. You claimed the opposite. You don't even know what you're talking about.

Oh yes I did several times now.

Nope, no mention of 40+fps before that. Again you don't know what you were talking about...

My assumption is, that you play 1920x1200 @ 30 fps AVC and compare it with 1920x1080 @ 60 VP9, just because you think it is the same, because of same LABEL at the player.

But look on my table I gave you in my previous post.

This is comparison between 1920x1080 @ 60 H.264 AVC vs 1920x1080 @ 60 VP9.

The used video is this one:

https://www.youtube.com/watch?v=mbsR82lJKMs

Your assumption is wrong.
You're the one who's imagining/assuming things. I check the resolution and the format during playback. And I made the comparison with downloaded streams not in the web player.

I really don't care about the table, I'm not in the habit of memorizing stream codes. As I said I look at the video.

Edit:
Apparently my comparison is invalid, because potplayer's VP9 decoder seems to be flawed...

De-M-oN

1st January 2016, 23:17

It plays 1200p AVC. You claimed the opposite. You don't even know what you're talking about.

I perfectly fine know.

And again for you the thousand time:

AVC plays all up to 8k, when fps is BELOW 41 FPS

If FPS is HIGHER THAN 41FPS, THEN AVC is only available up to 1080p as 60fps streams and up to 8k as 30fps streams

There is NO 60FPS streams in AVC for 1440p, 4k

you got it now? Again: look at the table.

Nope, no mention of 40+fps before that. Again you don't know what you were talking about...

I did. Dont you even know the word HFR?

You're the one who's imagining/assuming things.
wrong. I check format ID and download each ID and compare them. Nothing can go wrong with it.

Apparently my comparison is invalid, because potplayer's VP9 decoder seems to be flawed...

great.

____

And this is the most easy way to find out which formats are available:

http://www.h3xed.com/blogmedia/youtube-info.php

which I posted already too. This accesses directly youtube and lists every available quality and at which Label on the player it plays and which Format ID it has.

And like you can see there, there is no HIGH FRAMERATE Streams ever available for 1440 and 4k coded in AVC, just VP9 ones.

Only 30 fps is coded into AVC higher than 1080p.

So when you upload a 4k60 video you will get this scheme:

VP9 qualities

1080,1440,4k at 60fps and 30fps variants.

AVC qualities:

1080 at 60fps and 30fps variants
1440, 4k at 30fps, no 60fps variants here.

foxyshadis

2nd January 2016, 09:33

No need to get heated, guys, I misunderstood too.

I wonder if this is new Youtube policy or if they're just transitioning right now.

De-M-oN

2nd January 2016, 11:55

H.264 was never available for 1440p60 and 4k60 streams.

That the label is replaced instead of being a seperate one if you are inbetween 2 resolutions was changed to this (long ago). I think they did it for transparency - so that if the user sees a 1440p label for example, that he then gets this resolution.

I like that though. Its like a gift for better looking 1080p by uploading 2048x1152 now. And 1080p label is more often used, even people which totally forget bitrate and think >1080p they dont benefit from with their 1080p monitor would be reached with this trick now.

De-M-oN

2nd February 2016, 00:54

4k (you need at least 1750 height to get it (for example 3200x1800 with 16:9 aspect ratio)

For 4k they have removed the bitrate limit. Its just CRF now. No matter how complex the video gets - it just raises the bitrate - but doesnt look worse.

This is even a native 3840x2160 one:

https://www.youtube.com/watch?v=_3q7GYO31Xk

And even that is AWESOME QUALITY at 4k.

Nintendo Maniac 64

2nd February 2016, 01:29

4k (you need at least 1750 height to get it (for example 3200x1800 with 16:9 aspect ratio).

Don't forget that, if your horizontal resolution qualifies for 4k, you don't need to worry about your vertical resolution at all.

Example (3840x1350):
https://www.youtube.com/watch?v=55_hcqDXDbI

De-M-oN

2nd February 2016, 01:31

Same for height. You need at least 1750 height to get 4k.

But for common aspect ratios you reach the height far earlier.

The opposite is I think only happening with smartphones.

Roco

21st March 2016, 17:29

I downloaded a recent 4K video and according to Mediainfo it still uses AVC. Also in YouTube's recomendations only h.264 is mentioned. So, was the vp9 codec already abandoned?

De-M-oN

21st March 2016, 17:31

of course not. You simply just downloaded the H.264 file instead of the VP9 one.

Roco

21st March 2016, 17:44

Here is another one -there is no option for VP9 -at least with Internet Download Manager:

http://i2.imageban.ru/out/2016/03/21/e95e5e66a571b3031e2ed5f682cbf998.png

De-M-oN

21st March 2016, 17:46

You should change the way you download the videos. It apparently can only download the h.264 videos of youtube.

Use JDownloader 2.

Roco

21st March 2016, 17:55

You should change the way you download the videos. It apparently can only download the h.264 videos of youtube.

Use JDownloader 2.Its webinstaller raises security warnings from ESET Nod32 Smart Security and blocks the site.
Any other (safer) way?

De-M-oN

21st March 2016, 17:58

It is safe. Just decline the adware at setup of the software (the usual)

If you see a Decline Button at setup, press it.

And always use custom install so that you can control what is installed (this is for setups in general)

LithosZA

21st March 2016, 19:46

http://i645.photobucket.com/albums/uu177/lithosza/foreign.png (http://s645.photobucket.com/user/lithosza/media/foreign.png.html)

You can also see the VP9 version with the SaveFrom script for the TamperMonkey add-on for Chrome.

Roco

21st March 2016, 20:34

It is safe. Just decline the adware at setup of the software (the usual)
There is no option to allow a blocked site other than disabling Nod32 or certain features -which I avoid, as I've seen that when it blocks something or raises a warning, it's usually right.
So I used the Firefox addon "YouTube Video and Audio Downloader" instead, and it works OK, thanks.

The reason I wanted to find out about VP9 was to see if a 4K TV with VP9 support is required for 4K playback (using the TV's browser), but apparently YouTube provides both AVC and VP9 and I don't think they will abandon AVC any time soon, so VP9 support doesn't seem necessary. I've done extensive testing on previous VP versions several years ago and in all cases x264/h.264 was significantly superior in efficiency (or picture quality). Since I've missed some episodes since then, has anyone done any serious comparisons between VP9 and AVC?
I wouldn't be surprised if AVC is still superior.

Roco

21st March 2016, 20:38

You can also see the VP9 version with the SaveFrom script for the TamperMonkey add-on for Chrome.Nicer interface but I'm not using Chrome, but I'll keep that in mind, thanks.

De-M-oN

21st March 2016, 21:25

There is no option to allow a blocked site other than disabling Nod32 or certain features -which I avoid, as I've seen that when it blocks something or raises a warning, it's usually right.

But its wrong.

I use JDownloader 2 since years. My PC still lives. imagine that :o
False Positives happen. Realize that.

Also:

Do you really think the software would have such huge userbase:

http://abload.de/img/desktop_2016_03_21_21vsblr.png

if it would be just a piece of malware?

I wouldn't be surprised if AVC is still superior.

Not really.

H.264 1080 vs VP9 1080:

http://screenshotcomparison.com/comparison/156406

Also 1440p60 and 4k60 is only offered as VP9.

Roco

21st March 2016, 22:59

But its wrong.

I use JDownloader 2 since years. My PC still lives. imagine that :o
False Positives happen. Realize that. The purpose of undesirable software varies, it might not destroy your PC, but serve as spyware, identity theft, etc so that's not proof that it's innocent.

Do you really think the software would have such huge userbase:

http://abload.de/img/desktop_2016_03_21_21vsblr.png

if it would be just a piece of malware? A spyware etc can be a very useful piece of software -usefulness will hep it spread. If what's doing is not apparent to the users, most will download it and use it for a long time. That doesn't prove it's harmless.

To be fair, also Nod32's warning and site-block doesn't prove it's harmful, either. It's also possible that Nod32 is wrong.

Not really.

H.264 1080 vs VP9 1080:

http://screenshotcomparison.com/comparison/156406
If the source video is provided so another person can repeat the encoding and see if this screenshot is BS or not, only then it would be considered as "proof". Proof requires transparency.

Also 1440p60 and 4k60 is only offered as VP9That's good.

Not to be misunderstood, personally I want VP9 to succeed as much as possible, as long as it remains open, so I can use it in future commercial projects without having to pay MPEGLA a premium.

De-M-oN

21st March 2016, 23:23

If the source video is provided so another person can repeat the encoding and see if this screenshot is BS or not, only then it would be considered as "proof". Proof requires transparency.

Its my video.

And VP9 coding is better. Thats a fact. Youtube also doesnt encode to VP9 on very small channels to save CPU. VP9 coding is very cpu intense and while my videos are H.264 ready within some hours, vp9 needs up to a day until its ready.
Visit my channel:
http://www.youtube.com/c/demon66

That VP9 looks by far better applies for every video. So you can check any video out to compare.
But it should be most noticable on this very complex Rise of the Triad video:

https://www.youtube.com/watch?v=REMEt6s37fs

especially above 6:30. The textures at above 6:30 are very hard to compress.
___

Hell dont you get it? JD2 is clean. The NOD32 alert comes from that stupid adware in the setup. Never heard about setups which have adware in it which you have to remove the checkmarks to avoid install? For example opencandy anyone? https://en.wikipedia.org/wiki/OpenCandy Its not opencandy for JD, but the same stuff.
Just dont install the adware like you do with every other setups which contain adware checkmarks. Thats nothing special about such setups. They're rather quite common. The developer gains money if they use adware in their setups. If the user has a strong enough brain he removes the checkmarks and he is fine.
You joined this forum 2005, so you have at least eleven years of computer experience. You should know adware setups already. For JD you press a Decline Button to avoid installing the adware.

Roco

22nd March 2016, 00:20

Its my video.

And VP9 coding is better. Thats a fact. Youtube also doesnt encode to VP9 on very small channels to save CPU. VP9 coding is very cpu intense and while my videos are H.264 ready within some hours, vp9 needs up to a day until its ready.
Visit my channel:
http://www.youtube.com/c/demon66

That VP9 looks by far better applies for every video. So you can check any video out to compare.
But it should be most noticable on this very complex Rise of the Triad video:

https://www.youtube.com/watch?v=REMEt6s37fs

especially above 6:30. The textures at above 6:30 are very hard to compress.I don't need "facts", I need proof. I'll be convinced when both the source and the encoding is provided for download -I probably need to do this myself... :rolleyes:

Hell dont you get it? JD2 is clean. The NOD32 alert comes from that stupid adware in the setup. Never heard about setups which have adware in it which you have to remove the checkmarks to avoid install? For example opencandy anyone? https://en.wikipedia.org/wiki/OpenCandy Its not opencandy for JD, but the same stuff.
Just dont install the adware like you do with every other setups which contain adware checkmarks. Thats nothing special about such setups. They're rather quite common. The developer gains money if they use adware in their setups. If the user has a strong enough brain he removes the checkmarks and he is fine.
You joined this forum 2005, so you have at least eleven years of computer experience. You should know adware setups already. For JD you press a Decline Button to avoid installing the adware.I do know about adware, Nod32 doesn't warn on all adware and/or on the same degree, which is what makes it suspicious. And as I said, it has blocked the access to the site with no option to "allow" it, and disabling the antivirus is out of the question. That's the price to pay when a developer uses the most stupid way to make money. I've solved my problem as I mentioned, so I don't care about JD, thanks anyway.

De-M-oN

22nd March 2016, 00:29

I don't need "facts", I need proof. I'll be convinced when both the source and the encoding is provided for download -I probably need to do this myself...

is it so hard to go on my channel and download mp4 and vp9 of same quality level and compare?
Seems very hard job.
I record in 2560x1600 @ 50fps @ YV24 with MagicYUV Codec.
I encode with x264 @ CRF15 and I use YV24 for the encode as well. Is it good enough source for you? I hope so :rolleyes:

Here is the Mediainfo of the video capture of Legendary Part 13:

http://paste2.org/0XHwdthx

And here is the MeGUI Log of Legendary Part 13:

http://paste2.org/MHMPsJh7

But unfortunately I didnt consider that you'll say I just faked it. I'm sorry :(

You also seem to have absolutely no trust into VP9. How it comes that you think its a bad codec? Its actually between H.264 and H.265. Dont confuse it with the older VP8 which was indeed worse.
Also dont forget that youtube doesnt encode H.264 with preset slow. And of course no 2pass.
And do you really think youtube would invest a hell of a cpu load for VP9 and avoid encoding vp9 on small channels for the result of a worse quality? Thats an interesting logic. Makes no sense at all.
_
A bit pity about jd2. its a great program. But you seem to trust in heuristic hits more than anything. But okay.

foxyshadis

22nd March 2016, 10:50

I don't need "facts", I need proof. I'll be convinced when both the source and the encoding is provided for download -I probably need to do this myself... :rolleyes:

All the proof you need is to download any mp4 format from youtube and single-step it. The blocking artifacts will stand out in sharp relief; most of the time I can easily see them in real-time motion. In contrast, the VP9 both avoids blocking and typically has higher detail.

mzso

23rd March 2016, 11:00

I don't need "facts", I need proof. I'll be convinced when both the source and the encoding is provided for download -I probably need to do this myself... :rolleyes:

Then get it yourself and don't whine/spam about it. If someone were to provide the source/encode files. You could claim that they cheated, and dismiss it just like you did with the screenshots. Why would someone waste their time for you?

The reason I wanted to find out about VP9 was to see if a 4K TV with VP9 support is required for 4K playback (using the TV's browser), but apparently YouTube provides both AVC and VP9 and I don't think they will abandon AVC any time soon, so VP9 support doesn't seem necessary.

Youtube already doesn't provide high fps (50-60) variants for AVC videos. You'll only ever get max 30fps with AVC video.

Nintendo Maniac 64

24th March 2016, 19:55

Youtube already doesn't provide high fps (50-60) variants for AVC videos.

Uhh, actually they do; it's just that they have HFR AVC max out at 1080p; 1440p HFR and higher is only encoded into VP9.

(this actually makes sense since a good amount of AVC hardware decoders can't handle HFR at higher-than-1080p resolutions)

mzso

24th March 2016, 20:04

Uhh, actually they do; it's just that they have HFR AVC max out at 1080p; 1440p HFR and higher is only encoded into VP9.

(this actually makes sense since a good amount of AVC hardware decoders can't handle HFR at higher-than-1080p resolutions)

I forgot to add above 1080p. Still they're missing. Plus AVC is not as cpu heavy as VP9

Nintendo Maniac 64

24th March 2016, 20:31

I forgot to add above 1080p. Still they're missing. Plus AVC is not as cpu heavy as VP9

The issue is that web browsers default to using hardware decoding if it's available. You see, such hardware AVC decoders will in fact attempt to decode the likes of 1440p 60fps, but what will happen is that they will either have frameskips or the video will play in slower-than-realtime.

Besides, the majority of hardware decoders capable of HFR AVC at resolutions greater than 1080p are those that support some sort of hardware-accelerated VP9 decoding anyway.

There are of course exceptions - Intel GPUs since Ivy Bridge can decode AVC @ 2160p 60fps which completely predates HEVC and VP9, Nvidia since 2011 can decode AVC @ 2160p 30fps which is good enough for AVC @ 1440p 60fps, and modern AMD GPUs completely lack VP9 hardware acceleration despite having support for HEVC and AVC @ 2160p 60fps (maybe related to the failure of 20nm? It did result in the lack of HDMI 2.0 and possibly DisplayPort 1.3).

Roco

25th March 2016, 00:03

is it so hard to go on my channel and download mp4 and vp9 of same quality level and compare?
Seems very hard job.
....
I encode with x264 @ CRF15 and I use YV24 for the encode as well. Is it good enough source for you? I hope so
All the proof you need is to download any mp4 format from youtube and single-step it. The blocking artifacts will stand out in sharp relief; most of the time I can easily see them in real-time motion. In contrast, the VP9 both avoids blocking and typically has higher detail.
Then get it yourself and don't whine/spam about it. If someone were to provide the source/encode files. You could claim that they cheated, and dismiss it just like you did with the screenshots. Why would someone waste their time for you?
I neither wine, nor demand anything, from anyone and I'm not saying that VP9 is worse.

I just argue on your illogical claim that I should accept someone else's encoding and conclusions as proof!

Contrary to belief, scientific proof requires repeatability i.e to give the option to someone else to repeat what you did, in order to check and verify your claims. In this case that would be to try encoding myself both in x264 and VP9 a specific source clip you either suggest or provide (in order to compare the results).

I know the source file size would be huge -but a short sample of a few seconds might be manageable -even if it was 10 GBs I would download it, which would be ideal for all of us to have the exact same source and compare our findings, otherwise, if I play the same game (which I don't even own), the actual video will never be exactly as yours. A "demo" mode of that game's trial could be an alternative but I doubt that my dual-core CPU would manage to record the game without skipping frames.

There are many settings and methods evolved in the encoding of a video as you know, so there is a chance I'll achieve less artifacts than you. Or not. YouTube also might not doing it right for one reason or another.

zerowalker

25th March 2016, 00:11

If you want you can use my application i built in C#, a sort of simple GUI for youtube-dl.

//It contains ffmpeg as well as that's used for muxing if you want to download Video and Audio.
https://www.sendspace.com/file/yvgy6u

If you are afraid of using it i can instead give you the source code so you can compile it yourself.

(It's just a simple tool for my own needs, i know it's not well made;P!)

There are 3 text boxes.

First is for the URL, the two at the bottom are for your Username and Password if the video you want to download is a private one this must be used.
Otherwise just using the URL and pressing Get Format will give you all the formats, and you can select Video and Audio, separate or both and download them:)

De-M-oN

25th March 2016, 00:14

@Roco : errrr I have a slight feeling that we talk a bit (hmm how you would write in english? ) I mean into different directions.

I mean the results at youtube, and you seem to talk about local differences.

And fact is, that youtube's VP9 encodes look better than their mp4 ones. You can check it yourself with any video you have on your own. You dont need my video to check it. Upload any of your videos to youtube and see yourself which one looks better vp9 or mp4. And it will be vp9.

I would assume that VP9 wins over x264 also on a local encode, especially on the higher presets, but VP9 encoding will need much more time too.

Roco

25th March 2016, 00:29

@zerowalker Thanks very much, but I've already solved that problem with the firefox addon "YouTube Video and Audio Downloader".

errrr I have a slight feeling that we talk a bit (hmm how you would write in english? ) I mean into different directions.
I mean the results at youtube, and you seem to talk about local differences.So it was a misunderstanding, I was talking about which compression algorithm is more efficient (in terms of output size and picture quality) independently of a specific service like YouTube. My guess is that a proper local x264 encoding will result in much better results , but I might be wrong.

zerowalker

25th March 2016, 00:31

@zerowalker Thanks very much, but I've already solved that problem with the firefox addon "YouTube Video and Audio Downloader".

So it was a misunderstanding, I was talking about which compression algorithm is more efficient (in terms of output size and picture quality) independently of a specific service like YouTube. My guess is that a proper local encoding will result in less differences, but I might be wrong.

Oh, my prime moment as the messiah ended very quickly:(!
But awesome:)

Nintendo Maniac 64

25th March 2016, 01:53

My guess is that a proper local x264 encoding will result in much better results , but I might be wrong.

It largely depends on the bitrate. Basically the lower in bitrate you go, the better both VP9 and HEVC will do compared to x264.

mzso

27th July 2016, 11:57

What would you guys say? How much efficiency advantage does VP9 have in practice on youtube?
Should I get the AVC file if it's 5-10% larger? Or maybe it only makes sense if it's 15-20% larger? (I highly doubt that VP9 can be as good as a 30+% larger AVC stream).
The size difference can differ widely. Sometimes the AVC stream is more than double the size of the vp9 one, other times it's actually smaller a bit.

Did anyone do comprehensive tests?

zerowalker

27th July 2016, 11:59

I think VP9 is better upp to like 20%.
It's not that VP9 itself is better generally, it's just that their settings on AVC sucks compared to VP9.

In most cases VP9 will be the better choice i think, especially if the encoding is of this year i guess.

De-M-oN

27th July 2016, 12:13

On same quality level their VP9 videos look ALWAYS better than h.264.
You cant compare this with local encodes.
Youtube has to use fast encoding.
They spend more time into vp9 though.
But encoding needs significant more time and small channels dont get it either seldom, sometimes or even not at all.
Depends on the channel (size, views and many many other factors).

mzso

27th July 2016, 13:48

On same quality level their VP9 videos look ALWAYS better than h.264.

This is just false. As I said sometimes the filesize is half for VP9 videos and the quality is certainly does not reach that of the AVC stream.

De-M-oN

28th July 2016, 10:43

hmm really? I get a bit tired of you :(

We had this topic already, but there you misunderstood that I meant youtube and you thought I meant local encodes and I really thought this topic was done after this, but apparently not. This time you seem indeed to try to tell me that on youtube side VP9 looks worse than their h.264 encodes.

And if so I have to show you this comparison I gave you already back then again:

This is a comparison of their 1080p Quality Level

http://screenshotcomparison.com/comparison/156406

Yes. VP9 is certainly worse http://www.ugly-smilies.de/data/uglyconfused2.gif

I can show you any video, it will be always the same difference.

Also, honestly, ask yourself: Why should google invest a shitload on cpu for VP9 encoding, even due to this they give not every new channel directly the vp9 encodes, just because to reach worse quality than the current fast encodable h.264 videos? That wouldnt make any sense for google at all lol. Ask yourself if THAT would make any slightly sense for google.

And on the higher quality levels (1440p, 4k) the bitrate is even higher than the H.264 ones. 1080p is a bit lower though yes. But still looks BY FAR better than their h.264 crap encodes like you can see on the comparison.

mzso

28th July 2016, 11:30

You said: "VP9 videos look ALWAYS better than h.264" Which is clearly false. Sometimes it has half the bitrate for the same quality, it's certainly not better then.

You keep coming with the same nonsense. A single snapshot won't make you less wrong.
Also if memory serves, that screenshot is from a 60p video, in which case the results are likely different...

"This time you seem indeed to try to tell me that on youtube side VP9 looks worse than their h.264 encodes"
That's nonsense you're trying to put in my mouth... Stop strawmaning. You keep generalizing. Can't you think in anything other than (all) VP9 vs. AVC? How hard it is to comprehend that the actual bitrate varies greatly between the videos? Seriously...

mzso

28th July 2016, 11:33

De-M-oN

28th July 2016, 11:43

You keep coming with the same nonsense. A single snapshot won't make you less wrong.

The whole video looks worse in AVC. As if only this single frame would look worse xD
It applies for 720p30 and 720p60, it applies for 1080p30 and 1080p60
It is always worse.

Show me a comparison of a case where the opposite is the case. I would be surprised then.

The encoding by them is always the same for both.

At VP9 they use Constrained Quality (which is CRF with a bitrate upperlimit - pretty much like x264 with CRF+VBV combo (which btw is used at their h.264 videos)

thats why the bitrates vary.

But for example 1440p HFR VP9 the average bitrate upper limit is set at 15 mbit

4k VP9 even seems only CRF without a upperlimit, because even very complex videos got just higher average bitrate and the quality the same. And the CRF on this level is surprisingly good.

I can give you any video where the VP9 variants look much better

I give you a random video by me which I even didnt checked:

https://www.youtube.com/watch?v=43v83ExvbdE

The result will be the same. VP9 better than h.264.

HFR or non-HFR doesnt make the fact different.

De-M-oN

28th July 2016, 11:49

I also must admit that I'm so frustrated and so damn disappointed about their h.264 Encodes that I dont publish any video until VP9 is ready. Because I'm absolutely not satisfied with their h.264 encodes. They look just sooo bad..

Nintendo Maniac 64

28th July 2016, 12:09

I'd just like to mention that, in my experience, you simply can't get a good-looking video with the bitrate they use for AVC. Am I wrong?

De-M-oN

28th July 2016, 12:15

You could, but their encoding settings dont allow that. Google has a ton of traffic, but they care for cpu. They also could raise the bitrates for the lower qualities, but the average internet is still too bad so that they dont do. They said it like this on reddit.

Nintendo Maniac 64

29th July 2016, 01:11

You could, but their encoding settings dont allow that. Google has a ton of traffic, but they care for cpu.
But that's just it. When I tried it myself a years or so ago with crazy high quality settings at the very same bitrate, the results weren't really all that much better.

They also could raise the bitrates for the lower qualities, but the average internet is still too bad so that they dont do. They said it like this on reddit.

And thank goodness because I'm one of those people with not-great internet (3Mbps down).

mzso

19th October 2016, 14:16

Has anyone also notice a new trend of VP9 streams being almost the same size as their AVC counterparts? I'm hoping it's not the AVC ones that shrunk but the other way around.
There used to be a huge difference. (usually the AVC streams being much larger) Now it looks much clearer which one is worth downloading.

De-M-oN

19th October 2016, 14:36

The encoder werent changed in all quality levels for both encoders.

Remember that it uses not CBR or VBR fixed. It depends on complexity as well. Upload a desktop capture and you'll be surprised how low the bitrates will be on both encoders.

What I meant is: It is CRF with an average bitrate limit like described here : http://wiki.webmproject.org/ffmpeg/vp9-encoding-guide at constrained quality.

Now it looks much clearer which one is worth downloading

That was clear the whole time.

Nintendo Maniac 64

20th October 2016, 01:07

Well if I still had access to fast internet overnight, I'd do another one of my quality tests...

Maybe I could instead just have YouTube re-process one of my older test clips...though I've only done it three times where I uploaded at 1080p in lossless.

Perhaps in the future, before AV1 actually becomes a thing, I should use something like a 15 second video clip rather than a 4 minute clip.

NikosD

23rd October 2016, 20:31

It seems that latest Chrome x64 v54 when you write chrome://gpu at the address bar, it says VPx Video Decode: Hardware accelerated for Nvidia and AMD cards.

I don't have an Nvidia card capable of VP9 Hardware acceleration, maybe someone with a GTX 950/960 or Pascal card could try some Youtube VP9 videos.

Also, I don't know if previous versions of Chrome said the same thing.

My Intel system has VP9 acceleration disabled due to a Kabylake VP9 problem with HW acceleration.
You can read more here about Kaby and Chrome:
https://bugs.chromium.org/p/chromium/issues/detail?id=654111

Of course my Haswell iGPU even if VP9 acceleration was enabled, it wouldn't accelerate VP9 codec.

The activation of VP9 acceleration on AMD cards is meaningless because no card is capable of VP9 hardware decode, but it's good that it's there.

So, could anyone with an Nvidia card give a try with the new Chrome x64 and VP9 video from Youtube ?

De-M-oN

23rd October 2016, 20:44

VPx decoding isn't supported before Windows 10 anniversary update.: 616318
Disabled Features: accelerated_vpx_decode

I have a GTX 1080, but I hate Windows 10, I'll stay on Windows 7 as long as possible..

NikosD

23rd October 2016, 20:49

Right. I forgot to say what my signature says.

I always put latest MS OS on my system.

Nintendo Maniac 64

24th October 2016, 02:49

The activation of VP9 acceleration on AMD cards is meaningless because no card is capable of VP9 hardware decode

Not entirely, AMD does have a hybrid GPU accelerated VP9 decoder - it's just not full hardware VP9 decoding.

NikosD

24th October 2016, 04:18

Which AMD card has enabled hybrid VP9 acceleration ?

I know none.

Nintendo Maniac 64

25th October 2016, 05:05

Which AMD card has enabled hybrid VP9 acceleration ?

Polaris should:

http://www.amd.com/en-us/press-releases/Pages/amd-introduces-new-2016jul25.aspx[/url]"]Based on the Polaris architecture, the new Radeon Pro WX Series graphics cards employ the latest advances in AMD’s 4th generation Graphics Core Next architecture, uncompromising 4K video decode and encode support including HEVC2 and VP9 and next generation display support including DisplayPort 1.4-HDR3 and HDMI™ 2.0

http://www.amd.com/en-us/press-releases/Pages/radeon-rx-480-2016jun29.aspx[/url]"]Radeon™ RX 480 features AMD's most versatile decode accelerator ever, enabling beautiful HDR movies using 10-bit HEVC8 from the likes of Netflix and Amazon, accelerated H.264 video of streamers on Twitch, accelerated VP9 videos on sites like YouTube, and razor-sharp video calling using HEVC and MJPEG in applications like Skype.

NikosD

25th October 2016, 05:21

Polaris was released on the end of June.

After 4 months and still no VP9 acceleration has been enabled.

Even worse, the H.264 hw acceleration is broken in some cases.

I don't think we will see VP9 acceleration of Polaris soonish

sneaker_ger

8th November 2016, 09:34

Youtube just announced HDR availability starting today. Haven't seen word of the codec yet. It's for "supported devices", among others the Samsung SUHD 2016 (in the future) and the Chromecast Ultra. Does anyone know if the Chromecast Ultra supports VP9 10bit?
https://youtube.googleblog.com/2016/11/true-colors-adding-support-for-hdr.html

Jamaika

8th November 2016, 10:13

Very interesting message. I have another question about the HDR codecs.
How can I play videos HDR monitor Eizo EV2450 8bit?
I have a movie yuv422p10le BT2020 full range. When I play in ffplay probably turns on HDR, although the information HDR doesn't display. How to do it in MPC-HC plus madVR?

I'm surprised, too, that YouTube needs additional information in a container and not in the the codec. Currently, none of this information is reading: naither MPC player, neither the mediaInfo. :(

Was this information only to youtube with new TV?
mkvmerge \
-o output.mkv\
--colour-matrix 0:9 \
--colour-range 0:1 \
--colour-transfer-characteristics 0:16 \
--colour-primaries 0:9 \
--max-content-light 0:1000 \
--max-frame-light 0:300 \
--max-luminance 0:1000 \
--min-luminance 0:0.01 \
--chromaticity-coordinates 0:0.68,0.32,0.265,0.690,0.15,0.06 \
--white-colour-coordinates 0:0.3127,0.3290 \
input.mov

nevcairiel

8th November 2016, 10:37

Android defined a variety of HDR formats to be supported:
https://source.android.com/devices/tech/display/hdr.html

Two of those are VP9 Profile2, ie. VP9 10-bit, so unless YouTube streams HEVC HDR10 to the Chromecast Ultra, its likely it does support VP9 10-bit decoding.

The reason the vp9-pq mode puts the HDR information into the container is that the VP9 codec did not specify generic information signaling like H264/HEVC SEIs that can be arbitrarily extended.
Also by using the container, YouTube can accept a variety of codecs and not be limited to HEVC with the built-in metadata.

sneaker_ger

8th November 2016, 10:53

Ok, youtube-dl already shows the streams (https://www.youtube.com/watch?v=tO01J-M3g0U):
[youtube] tO01J-M3g0U: Downloading webpage
[youtube] tO01J-M3g0U: Downloading video info webpage
[youtube] tO01J-M3g0U: Extracting video information
[youtube] tO01J-M3g0U: Downloading MPD manifest
[info] Available formats for tO01J-M3g0U:
format code extension resolution note
249 webm audio only DASH audio 50k , opus @ 50k (48000Hz), 912.10KiB
250 webm audio only DASH audio 65k , opus @ 70k (48000Hz), 1.15MiB
140 m4a audio only DASH audio 127k , m4a_dash container, mp4a.40.2@128k (44100Hz), 2.34MiB
251 webm audio only DASH audio 130k , opus @160k (48000Hz), 2.30MiB
171 webm audio only DASH audio 131k , vorbis@128k (44100Hz), 2.07MiB
160 mp4 256x144 DASH video 115k , avc1.4d400c, 30fps, video only, 2.05MiB
278 webm 256x144 DASH video 122k , webm container, vp9, 30fps, video only, 1.69MiB
133 mp4 426x240 DASH video 261k , avc1.4d4015, 30fps, video only, 4.53MiB
242 webm 426x240 DASH video 272k , vp9, 30fps, video only, 3.61MiB
243 webm 640x360 DASH video 513k , vp9, 30fps, video only, 6.92MiB
134 mp4 640x360 DASH video 663k , avc1.4d401e, 30fps, video only, 9.09MiB
244 webm 854x480 DASH video 988k , vp9, 30fps, video only, 12.66MiB
135 mp4 854x480 DASH video 1223k , avc1.4d401f, 30fps, video only, 17.87MiB
247 webm 1280x720 720p 2036k , vp9, 30fps, video only, 25.57MiB
136 mp4 1280x720 720p 2432k , avc1.4d401f, 30fps, video only, 34.81MiB
248 webm 1920x1080 1080p 3260k , vp9, 30fps, video only, 44.89MiB
302 webm 1280x720 DASH video 3261k , vp9, 60fps, video only, 41.13MiB
298 mp4 1280x720 DASH video 3495k , avc1.4d4020, 60fps, video only, 47.53MiB
137 mp4 1920x1080 1080p 4463k , avc1.640028, 30fps, video only, 65.31MiB
303 webm 1920x1080 DASH video 5108k , vp9, 60fps, video only, 71.01MiB
299 mp4 1920x1080 DASH video 5819k , avc1.64002a, 60fps, video only, 84.46MiB
264 mp4 2560x1440 DASH video 10437k , avc1.640032, 30fps, video only, 146.87MiB
271 webm 2560x1440 1440p 11064k , vp9, 30fps, video only, 134.73MiB
308 webm 2560x1440 DASH video 15731k , vp9, 60fps, video only, 204.25MiB
313 webm 3840x2160 2160p 21981k , vp9, 30fps, video only, 314.26MiB
266 mp4 3840x2160 DASH video 23340k , avc1.640033, 30fps, video only, 286.65MiB
315 webm 3840x2160 DASH video 29805k , vp9, 60fps, video only, 458.12MiB
330 webm 256x144 144p60 156k , vp9.2, 60fps, video only, 2.38MiB
331 webm 426x240 240p60 256k , vp9.2, 60fps, video only, 3.87MiB
332 webm 640x360 360p60 485k , vp9.2, 60fps, video only, 7.35MiB
333 webm 854x480 480p60 909k , vp9.2, 60fps, video only, 13.83MiB
334 webm 1280x720 720p60 1991k , vp9.2, 60fps, video only, 28.18MiB
335 webm 1920x1080 1080p60 3201k , vp9.2, 60fps, video only, 49.70MiB
336 webm 2560x1440 1440p60 11166k , vp9.2, 60fps, video only, 170.23MiB
337 webm 3840x2160 2160p60 20122k , vp9.2, 60fps, video only, 335.45MiB
17 3gp 176x144 small , mp4v.20.3, mp4a.40.2@ 24k
36 3gp 320x180 small , mp4v.20.3, mp4a.40.2
43 webm 640x360 medium , vp8.0, vorbis@128k
18 mp4 640x360 medium , avc1.42001E, mp4a.40.2@ 96k
22 mp4 1280x720 hd720 , avc1.64001F, mp4a.40.2@192k (best)
It's VP9 10 bit from 144p to 4Kp60 (streams 330 to 337).

NikosD

8th November 2016, 10:57

Interesting...
For VP9 10bit decoding and 4Kp60 resolutions, you will definitely need a HW decoder which means only Kabylake.

sneaker_ger

8th November 2016, 11:09

The reason the vp9-pq mode puts the HDR information into the container is that the VP9 codec did not specify generic information signaling like H264/HEVC SEIs that can be arbitrarily extended.
Also by using the container, YouTube can accept a variety of codecs and not be limited to HEVC with the built-in metadata.
Indeed, mkvinfo shows HDR container metadata:
| + Codec ID: V_VP9
| + Video track
| + Pixel width: 3840
| + Pixel height: 2160
| + Video colour information
| + Colour matrix: 9
| + Colour range: 1
| + Colour transfer: 16
| + Colour primaries: 9
| + Video colour mastering metadata
| + Max luminance: 1000
| + Min luminance: 0.001
| + Red colour coordinate x: 0.68
| + Red colour coordinate y: 0.31996
| + Green colour coordinate x: 0.26494
| + Green colour coordinate y: 0.68996
| + Blue colour coordinate x: 0.15
| + Blue colour coordinate y: 0.05998
| + White colour coordinate x: 0.3127
| + White colour coordinate y: 0.32896

Colour matrix: 9 = BT2020 Non-constant Luminance
Colour range: 1 = TV Range
Colour transfer: 16 = SMPTE ST 2084
Colour primaries: 9 = ITU-R BT.2020

Jamaika

8th November 2016, 11:16

Interesting...
For VP9 10bit decoding and 4Kp60 resolutions, you will definitely need a HW decoder which means only Kabylake.
No. Don't need a HW decoder. Only TV 4K HEVC/VP9 10bit + Chromecast Ultra.
With Chromecast and Chromecast Ultra, you use the same familiar mobile apps you already know and love — now with your TV. Just tap the Cast button in your favorite mobile app to start watching on TV.

Chromecast Ultra supports 4K, HDR and Dolby Vision, so you’ll get a crisper picture with higher resolution and more vibrant colors. At first, you'll be able to stream 4K content from Netflix, YouTube and Vudu, and we’re working to bring more 4K and HDR content on board. Later this year, Google Play Movies & TV will be rolling out 4K content. Even if your TV isn’t 4K-ready, Chromecast Ultra automatically optimizes your TV’s picture.

Mirror any content from your Android device or from a browser tab on your laptop (using Chrome) to the TV. And it comes in a small size that hides nicely behind your TV, without adding clutter to your living room.

Chromecast Ultra will be available in November for $69 from Best Buy, the Google Store, Target, and Walmart in the U.S., as well as from international retailers in 15 more countries.

So google Chromecast Ultra HDR doesn't include Europe (EU). It is not designed in desktops and laptops.

sneaker_ger

8th November 2016, 11:26

Chromecast Ultra will be released in Europe as well, you can already order it from e.g. the "German" Google Store. And we can assume it won't do 4K software decoding, i.e. NikosD's point about hardware decoding is perfectly valid.

No. Don't need a HW decoder. Only TV 4K HEVC/VP9 10bit + Chromecast Ultra.
If you have a Chromecast Ultra with the required decoder then you obviously won't need a TV with HEVC/VP9 decoder. Can't decode twice. Samsung SUHD 2016 won't need any Chromecast.

Jamaika

8th November 2016, 11:37

If you have a Chromecast Ultra with the required decoder then you obviously won't need a TV with HEVC/VP9 decoder. Can't decode twice. Samsung SUHD 2016 won't need any Chromecast.
OK. What kind of monitor do I have in HDR? 8 or 10 bit depths.

Edit: As it is with the HDR???

Colour range: 1 = TV Range
https://img.global.news.samsung.com/global/wp-content/uploads/2016/05/HDR1_TV_Main_1.jpg

Nintendo Maniac 64

9th November 2016, 01:48

OK. What kind of monitor do I have in HDR? 8 or 10 bit depths.

You technically can do HDR with only 8bit but you'd get ugly banding, so all the HDR standards use at least 10bit.

De-M-oN

9th November 2016, 01:51

Why no 10bit encodes without that bt.2020 stuff??

10bit would be useful for everyone (except mobile devices I assume). So by far more people..

I would love to see 10bit encodes on youtube. 8bit looks bad compressed on darkness. so much banding especially on lower bitrates, which 10bit really helps out a lot...

Nintendo Maniac 64

10th November 2016, 01:08

I would love to see 10bit encodes on youtube.

It's be much more likely for YouTube to use 10bit for non-HDR stuff when they introduce AV1 since it'll support 10bit from the beginning.

De-M-oN

10th November 2016, 01:11

AV1? Whats that? Apparently I missed an info? Clear me up please :)

david.copperfield1337

10th November 2016, 02:11

AV1? Whats that? Apparently I missed an info? Clear me up please :)
http://forum.doom9.org/showthread.php?t=172550

Jamaika

10th November 2016, 09:45

He doesn't understand certain things.
FFmpeg always suggests that the codecs VP9 & AV1 only color range tv.
According to me this is not true. The output files have different sizes. The color range PC is properly larger.
Stream #0:0: Video: vp9 (Profile 3) (VP90 / 0x30395056), yuv422p10le(tv, bt2020nc/unknown/unknown), 1920x1080, 29.97 tbr, 29.97 tbn, 29.97 tbc
ffmpeg.exe -loglevel verbose -i "yabadu20.avi" -frames:v 500 -an -f yuv4mpegpipe -vf scale=1920:1080:in_color_matrix=rgb:in_range=full:out_color_matrix=bt2020_ncl:out_range=pc/tv,format=yuv422p10le -strict -1 - |
vpxenc.exe -v --i422 --bit-depth=10 --input-bit-depth=10 --codec=vp9 --pass=1 --passes=1 --good --threads=4 --drop-frame=100 --end-usage=q --profile=3 --cpu-used=1 --min-q=5 --kf-max-dist=60 --auto-alt-ref=1 --cq-level=24
--target-bitrate=0 --aq-mode=1 --frame-boost=1 --color-space=bt2020 --ivf --output="vp9_rangefull.ivf" -
File size 500 frames:
Color range pc: 41,7MB
Color range tv: 36,1MB
Nothing improves command in the container webm.
mkvmerge.exe --colour-matrix 0:10 --colour-range 0:2 vp9_rangefull.ivf --webm -o vp9_rangefull.webm
PS If I'm wrong, please correct.

RBG

10th December 2016, 11:55

Hi.
Sorry if this is not an appropriate thread or my question was already answered, but can somebody explain me why VPx hardware decoder is blocked in Google Chrome for all operating systems lower than windows 10 anniversary update? Recently I bought a Pascal GPU(1060 in particular) and I thought that at last I will get VP9 hw decoding on youtube, because 1060 obviously supports it. But on my system (Windows 7 SP1 x64 + updates) Google Chrome turns off VP9 hw decode for some reason. I also checked DXVA checker log and as far as remember there definitely was a string regarding VP9 vld decoding(I can't check it now because I RMAed my GPU due to terrible coil whine), so it means that it is supported by the system, but Chrome thinks otherwise.

Here is what I found in chrome://gpu.
Graphics Feature Status
*
*
*
*
VPx Video Decode: Software only, hardware acceleration unavailable

Problems Detected
VPx decoding isn't supported before Windows 10 anniversary update.: 616318 (https://bugs.chromium.org/p/chromium/issues/detail?id=616318)
Disabled Features: accelerated_vpx_decode

NikosD

10th December 2016, 22:01

De-M-oN

10th December 2016, 22:04

if they would've made a good windows, they wouldnt have to force people switching from windows 7.

RBG

10th December 2016, 23:18

It's not only Nvidia's decoder that is blocked.

It's a decision of Google to not enable any hw VP9 decoding prior to Win 10 anniversary, probably because MS wanted to.

So it's actually Microsoft in collaboration with Google.

Nothing personal to Nvidia.

Well It doesn't look like Microsoft really wanted or pushed this idea because VP9 decoding is still accessible via DXVA API and other browsers can still use it, at least in theory. It seems that it is Google who doesn't want to spend additional money to support VP9 decoding on several platforms or to support several ways to access VP9 hw decoder. I mean not only they removed VP9 hw decoding support on Windows lower 10 RS1, but they also cut out Intel VP9/VP8 MFT support for no reason.

leonccyiu

9th January 2017, 13:17

I don't know if other people notice this but it seems like for new 4k videos, Youtube only encodes the 4k version in VP9. I can understand the decision, but for those using Safari on a Mac or Microsoft Edge without hardware acceleration for VP9, they won't be able to get the 4k version.

De-M-oN

9th January 2017, 13:27

Thats nothing new. Was always like this.
1440p HFR and 4k HFR were always encoded just into VP9.

Only 30fps videos are offered as h.264 as well on these resolutions.

leonccyiu

9th January 2017, 19:46

Thats nothing new. Was always like this.
1440p HFR and 4k HFR were always encoded just into VP9.

Only 30fps videos are offered as h.264 as well on these resolutions.

Not even 24/25/30 fps versions are offered in 4k now. I know that 4k60 and 1440p60 was never offered in h264.

sneaker_ger

9th January 2017, 19:52

I cannot confirm. This (https://www.youtube.com/watch?v=Oj9_ulp_AO0) was uploaded today and has 2160p24 available as AVC.

leonccyiu

13th January 2017, 02:38

http://wccftech.com/youtube-longer-supports-4k-safari/
https://www.reddit.com/r/apple/comments/5ixweg/since_when_there_is_no_4k_option_for_youtube_in/

When I use 4k download, there does appear to be a h264 version but in potplayer, and on firefox, there's only 1440p for h264.

CruNcher

13th January 2017, 18:05

Youtube does whatever fits the device ecosystem the best the target overall is maximum reach.

And you can force youtube to deliver to you whatever you want manipulating the client.

VP9 is meanwhile delivered through the whole range from lowest up to 4K

Most probably openh264 would fail 4k playback on your configuration via Firefox

On Safari either the internal decoder is not powerful enough or no hw support available.

Indeed the logic behind these decissions what gets delivered to whom and why and what not is tricky ;)

Though no one want's that you get delivered something where playback efficiency would be bad neither the Browser Vendor nor Youtube want that ;)

But the biggest problem are mostly of political nature ;)

In case of Apple they mostly decide based on the actual system specs what happens and what gets delivered and why for maximum user satisfaction, though this currently clashes with some things of political nature.

chummy

31st January 2017, 11:26

I asked about it in another topic but since no one give attention i will try here where some users like uploading to Youtube. Some time ago another guy noticed issue with NVENC H264 and Youtube reencoding. That guy said to me than when uploading NVENC source to Youtube the reencoding results were worse than feeding with x264 source.

Example i recorded gameplay with NVENC 50Mbit, created a second file reencoded with x264 slow preset at 30Mbit. Then i uploaded both videos to Youtube and x264 source reencodes always better, samething happens when using a common x264 encoder software. I tried samething with Quicksync and find it happens same way. Apparently when you reencode ASIC source in x264 with too low bitrate(1080p@5mbit) it give crappy results, but using good amount of bitrate in the reecndoing for example the 30mbit file i did will not notice that big difference. I just dont know why this happens and if people here already knows that and can share their point on it.

De-M-oN

31st January 2017, 15:45

NVEnc is even with 130 mbit a heavy loss of quality already on the local file.

chummy

1st February 2017, 11:15

But the point is than even you firstly reencoding from NVENC 50mbit source with x264 at 20Mbit which will give some few more loss over 50mbit NVENC source and after that reencode both files with x264 at 5mbit the 20mbit x264 source will generate a better output file.

Then the point why reencoding at low bitrates from NVENC source produce worse results than from a x264 source which has lower quality.

De-M-oN

1st February 2017, 13:57

20 Mbit x264 will look better than 50 mbit NVEnc.
Codecs and it's encoders have different compression efficiency.

dapperdan

1st February 2017, 16:58

It's confusing but the way I'm understanding it is they claim that NVenc encoded with x264 and then uploaded to Youtube looks better than the same NVEnc file uploaded directly to Youtube without the intermediate re-encode step. Which is odd, if that is the case, since you'd expect the extra generation to degrade things.

Do you have two examples to compare on Youtube?

De-M-oN

1st February 2017, 17:54

who says that multiple encodes would look better?

The best would be to upload lossless. Due to filesize problems we re-encode it. But the re-encode should be as good as possible (good encoder and highest filesize you can live to upload with) And of course a big resolution. 1080p will always look bad on youtube.
So the recommended way is to capture your games lossless, and not already lossy.

chummy

7th February 2017, 15:34

I just found out a 1080p30fps video which Youtube applied 6.7mbit/s for VP9 version against 4mbit/s for H264 version. What a strange move. The guy applied a heavy amount of sharp to video where it has some parts lots of vegetation plus high motion which make the video looks a big mess of macroblocks even in VP9 version in these parts. But the bitrate used for 1080p@30fps for VP9 is very far from the average they used of 2.5-3.5mbit/s and yet they used higher bitrate than for H264 version which is unprecendent for my experience.

De-M-oN

7th February 2017, 15:45

Never seen a 6,7 mbit 1080p video. Let alone 30fps.

How do you measure the bitrate?

sneaker_ger

7th February 2017, 16:14

"youtube-dl -F [youtube_video_url]" shows the sizes of the streams. Rest is math. (Or you download the streams and run MediaInfo on them)

De-M-oN

7th February 2017, 16:16

Use Mediainfo, the given values in youtube-dl are either inaccurate or I dont know the used unit. But if its bytes, it is inaccurate in terms of average bitrate.
So: Download it and lookup with Mediainfo to see the right values. And I assure you it wont be 6,7 mbit average.

sneaker_ger

7th February 2017, 16:24

Use Mediainfo, the given values in youtube-dl are either inaccurate or I dont know the used unit.
It literally spells out the unit as "MiB". And from what I can see the numbers are accurate down to 0.01 MiB.

chummy

7th February 2017, 16:35

Normally Youtube uses a different amount of bitrate for 30fps vs 48/60fps videos. H264 for example stuck at 4.2mbit/s and lower for 1080p30 and 5.8mbit and lower for 1080p48/60. For VP9 is samething but with lower average than H264 since for VP9 they are using very good compression presets against the H264 version.

In fact for my taste VP9 and HEVC looks better for gameplays with high details plus high motion since these codecs has better method to destroy details(sharp edges) despite it leave some ringing type artifact but this avoid heavy pixelization for hard content like H264 does.

De-M-oN

7th February 2017, 16:37

They use just 3,5 mbit on 1080p60 VP9 videos.. Chummy where do you lookup their bitrate?
And of course their VP9 videos have better quality.

chummy

7th February 2017, 16:41

The video is 1,11GB size and has 22min26sec. Make your maths.

De-M-oN

7th February 2017, 16:43

URL of the video?

chummy

7th February 2017, 16:46

I download them and check with mediainfo. I uploaded a 1440p@60fps video with only 37 seconds and the 1080p60@VP9 received 4.6mbit/s. Apparently when the full lenght of video is so hard to encode it pushes Youtube bitrate at it limits.

De-M-oN

7th February 2017, 16:50

This one has just 3,5 at 1080p.

https://www.youtube.com/watch?v=gRmSvqmio70

The street lost its texture. Just a flat grey.

chummy

7th February 2017, 17:31

If you want to look what a complex footage means look at that mess with almost 2x your bitrate.

https://youtu.be/wo0zxKVOJGg?t=317

The guy make it much worse because sharp, without it would look less blocky and only blurry.

De-M-oN

7th February 2017, 18:03

it would stay blocky I assume. Vegetation is always a problem. Even worse than my F1 video. Thats right.

But also the F1 video is already decent complex at the road.

sourcevideo:
http://www.abload.de/image.php?img=novaf1wmusa_qualifyinhtxqi.png

youtube:
http://www.abload.de/image.php?img=desktop_2017_01_19_18w7o4w.png

No direct image links so that you can compare them better due to different resolutions.

The street has no texture anymore and this is REALLY disappointing

Youtube also reduced some weeks ago the bitrates of ID 308 and ID 315, so that ID 315 has no quantizer compression anymore, now a cap of 15mbit. Before ID 315 was 99% like the source video. But now lossy as ...............
And ID 308 was reduced from 15 mbit to 10mbit...

CruNcher

7th February 2017, 19:48

It would be interesting to analyse if your bitrate meanwhile gets maybe somehow scaled on some commercial viable metrics like your clicks or overall add margin ;)

a good Game sequence test is also physical based rendering one at 60 FPS like the Endoor Speeder Scene of Star Wars Battlefront.

Thinking to get this kind of Temporal Motion stable decently at that low bitrates perceptually detailed is crazy overall, it's just able most of the time to go to the edge of the tires and then dies and even that looks shitty because it becomes like artifacting and is badly percepted overall by the viewer in Motion, it's overall better to eliminate all the details cohesive to reduce the bad feeling on the overall viewers Temporal perception, especially when you target Gamers ;)

And don't forget these textures are compressed already and mostly have artifacts, you could try to post process each Engine output beforehand understanding how the rendering works exactly and optimize for that but even then it wont save enough for that Temporal stability you aiming for :)

So you can't do a lot on your side.

Some already gave up ;)

And if Richard gives it up doing his own Service then that means something ;)

De-M-oN

7th February 2017, 19:50

I've send a 20k subs youtuber my F1 Video and it got exactly the same bitrates.

chummy

7th February 2017, 22:55

But your F1 video even with blurry road dont make it look astrocious like vegetation blocky/pixelization. F1 video yet has enjoyable look from viewer perspective, something i can't say about the video i linked.

Im not aiming to achieve perfect quality and save details at youtube 1080p under 5mbit, i just try to avoid the worst cases like the heavy macroblock pixelation for hard footage with vegetation in H264. Some games dont have blur or good AA and need some post processing like adding blur+AA filters to try decrease the issue with Youtube H264 1080p version or doing upscale. The good point in upscaling such videos(1440p) is than even the 1080p downscaled version will look better than the original 1080p without upscale, but i dont like upscaling.

Even if there is VP9 version Youtube still provide H264 version for some viewers since browser like Edge force Hardware acceleration by standard and majority hardware out there is H264 only. So i want to make my H264 avoid macroblock pixelation.

Another thing yet i dont understand is if there has a specific way for Youtube apply VP9, i have 2 videos with 2k+ views and even some with 500+ where they not applied VP9 yet. Some videos receive VP9 even at few views like 50 or 100. In past they always applied VP9 only to my videos which reach some higher number of views(around 1k), but that is not happening anymore. Now is very random.

Nintendo Maniac 64

8th February 2017, 01:18

I always liked to test with F-Zero GX since, even though it's textures are not high-res, it's quite a visually vibrant game that is very fast yet completely lacks motion blur.

chummy

9th February 2017, 08:16

Another 1080p@60fps video received special VP9 bitrate(7.2mbit/s), Youtube changing their standards?

https://www.youtube.com/watch?v=qX9k9XaBcAg

De-M-oN

9th February 2017, 18:04

Would be nice, but I assume my game videos will still get shit bitrates :(

chummy

9th February 2017, 18:38

Two videos i recently upped for test at 2048x1152p60fps not received 1152p60fps version, Youtube only created H264 1080p60 and 1152p30fps, so they're displaying the 1080p@60fps. For 1152p30fps upload there is not problem, they display it instead. For 1152p High framerate they only encode at VP9 like 1440p and 4k from what i see, then if you not receive special treatment with VP9 version you get stuck with 1080p.

Yet there is a benefit displaying a complex footage at 1080p which is a downscaled from 1152p or higher res upscale since it smooth out sharp edges and make the 1080p downscaled version get less artifacting/pixelation than an original 1080p with all crispy details.

CruNcher

10th February 2017, 12:30

I always liked to test with F-Zero GX since, even though it's textures are not high-res, it's quite a visually vibrant game that is very fast yet completely lacks motion blur.

Anyone knows a recent Touhou 60 fps upload ?

https://www.youtube.com/watch?v=jlW69GHniFg

hmm 720p 60

geez that looks really bad hmm it only switches to AVC no VP9 even as recently uploaded

Wonder how long it will take to be queued by the VP9 Encoder task if at all.

It gets automatically switched to the VP9 version

https://www.youtube.com/watch?v=qX9k9XaBcAg

no retranscode maybe ?

138 MB AVC
182 MB VP9

chummy

10th February 2017, 14:26

Ok, some hours yesterday after upload a non listed 1152p60fps video i decided to force Youtube reencode it to see if VP9 come out, so after almost 24 hours it happened. Now the video has VP9 version. Yet i have another non listed 1152p30fps video uploaded 31 january which still only has H264. I will try forcing reencoding to see what happens.

chummy

10th February 2017, 23:12

So i forced Youtube reencoding for the two videos with only H264 and now some hours later it received VP9 version. Interesting why it wont happen automatically at beginning. Some thought about it?

De-M-oN

11th February 2017, 00:20

I get it directly every time. But the bitrate changes are so frustrating... :(
Especially that ID 315 is now limited to 15mbit while before being CRF or Constant Quantizer (one of the both) Now this damnit 15mbit cap which makes look every video horrible again -.-

CruNcher

12th February 2017, 01:54

So i forced Youtube reencoding for the two videos with only H264 and now some hours later it received VP9 version. Interesting why it wont happen automatically at beginning. Some thought about it?

It seems to be taking some time for recent uploads

https://www.youtube.com/watch?v=jlW69GHniFg

is now also VP9

Older ones get lower priority it seems but the question is from what do they actually get encoded the first 2nd Generation or the second 2nd Generation inputs.

Which then opens the question for how long does Youtube actually save the original master uploads ;)

sneaker_ger

12th February 2017, 02:56

There's no indication they ever delete the originals. When they introduced 60 fps it activated on uploads that were several years old.

chummy

12th February 2017, 03:27

This Touhou 8 gameplay has heavy blocky artifacts even at VP9, looking another video from this game with 480p30fps only has no block artifacts. I think for high motion gameplays it's better to keep lower fps recording for Youtube to improve quality.

chummy

20th February 2017, 00:51

Youtube encoder go beyond the average bitrate values when add sharp for complex footage, just upped a short clip with heavy sharp like the other video and Youtube encoded H264 1080p version at 7.6mbit/s. Yet sharp is pointless since its bitrate hungry and will cause more artifacting damage even at higher bitrates.

De-M-oN

20th February 2017, 01:22

upload a video full of noise and the bitrate will explode. But as long as we dont get it on normal videos it helps nothing like you said^^

chummy

22nd February 2017, 21:03

Upped same video to Vimeo to test first time, they encoded my 1080p30 at 4.5mbit/s with 2 pass:

Encoding settings: cabac=1 / ref=4 / deblock=1:0:0 / analyse=0x3:0x113 / me=hex / subme=8 / psy=1 / psy_rd=1.00:0.00 / mixed_ref=1 / me_range=16 / chroma_me=1 / trellis=2 / 8x8dct=1 / cqm=0 / deadzone=21,11 / fast_pskip=1 / chroma_qp_offset=-2 / threads=34 / lookahead_threads=5 / sliced_threads=0 / nr=0 / decimate=1 / interlaced=0 / bluray_compat=0 / stitchable=1 / constrained_intra=0 / bframes=3 / b_pyramid=2 / b_adapt=1 / b_bias=0 / direct=3 / weightb=1 / open_gop=0 / weightp=2 / keyint=infinite / keyint_min=30 / scenecut=40 / intra_refresh=0 / rc_lookahead=50 / rc=2pass / mbtree=1 / bitrate=4500 / ratetol=1.0 / qcomp=0.60 / qpmin=5 / qpmax=69 / qpstep=4 / cplxblur=20.0 / qblur=0.5 / vbv_maxrate=4950 / vbv_bufsize=13500 / nal_hrd=none / filler=0 / ip_ratio=1.40 / aq=1:1.00

This same video has keyframe pumping(higher pixelation in keyframes) in Youtube with other P-B frames becoming "better", a common issue users find in Youtube with vegetation+motion. Now in Vimeo with their H264 settings the video is not causing issues at keyframes. Vimeo is doing 3 seconds GOP against 5 seconds from Youtube.

De-M-oN

22nd February 2017, 21:15

dailymotion encodes 1080p at 6mbit preset slow 2pass.

Nintendo Maniac 64

1st March 2017, 05:47

Remember how YouTube used to have their non-DASH AVC 720p encode be considerably higher quality than their DASH AVC 720p encode? And then a few months back they ditched the higher-quality encode for non-DASH and started using the same identical lower-quality encode for both DASH and non-DASH AVC 720p?

Well for some incredibly strange reason, the following video (which was not only uploaded just yesterday but also created yesterday) actually has that higher-quality encode for the non-DASH AVC 720p:

https://www.youtube.com/watch?v=ElV2Pvg0lx0

I have no idea why this is since no other video I checked is like that.

1-week-later EDIT: And now the higher-quality non-DASH AVC 720p encode has been replaced by the lower-quality AVC 720p encode that is also used for DASH.

lolwut YouTube.

chummy

20th May 2017, 15:08

Youtube is accepting HEVC uploads now, another guy informed about it.

NikosD

4th June 2017, 17:31

After Windows Creators Update, the old Chrome trick via the --enable switch to force enable VP9 hardware acceleration on Polaris RX 400/500 cards doesn't properly work.

Does anyone have any idea how to enable the VP9 acceleration of Chrome/YouTube on Polaris cards ?

I want access to VP9 4K60fps content, so no other browser is suitable for that.

Vincent Vega

8th June 2017, 23:59

in some of the youtube vp9.2 hdr videos (downloaded with youtube-dl -f 337 url) i noticed matrix specified as bt.709 while primaries bt.2020, examples:
https://www.youtube.com/watch?v=WW2DKBGCvEs
https://www.youtube.com/watch?v=5W_WVfxv6Ts
(althouh many others including those from the official 'HDR Launch Playlist' didn't suffer from such issue)

mkvmerge offers options:
--colour-matrix TID:9 for BT2020 non-constant luminance
--colour-matrix TID:10 for BT2020 constant luminance

i tried to remux the downloaded webm with both the above options.
the original and both remuxed versions play/look exactly the same (except madvr reports different matrices in OSD).

so which matrix is correct for such youtube videos?
why madvr outputs same picture no matter which matrix is reported by upstream?
is it still a good idea to change the matrix field in these videos to bt2020?

Nintendo Maniac 64

9th June 2017, 01:33

I want access to VP9 4K60fps content, so no other browser is suitable for that.

Is your Core i5-2400 not fast enough for software 4k 60fps VP9 decoding, even if it's in something like mpv or MPC-HC with D3D fullscreen?

NikosD

9th June 2017, 05:38

Is your Core i5-2400 not fast enough for software 4k 60fps VP9 decoding, even if it's in something like mpv or MPC-HC with D3D fullscreen?
If you mean software offline decoding (downloaded stream decoded locally from a HDD) then Core i5 is more than enough.

But online software decoding of 4K60fps VP9 using latest Chrome x64 is not feasible with Core i5.
I need faster CPU or HW decoding.

mzso

9th June 2017, 08:36

If you mean software offline decoding (downloaded stream decoded locally from a HDD) then Core i5 is more than enough.

But online software decoding of 4K60fps VP9 using latest Chrome x64 is not feasible with Core i5.
I need faster CPU or HW decoding.

That's rather sad. Isn't there a chromium fork with a better decoder?

I want access to VP9 4K60fps content, so no other browser is suitable for that.

This is just not true all. I just looked at an UHD 60fps video and it plays in VP9. Just tried it in firefox stable for your sake and you only need to flip media.mediasource.webm.enabled to true in about:config, and then you get served vp9 videos just like in Chrome. So you can give it a try if it's better. (Also FF 54 is released in a week, which may, or may not perform better)

sneaker_ger

9th June 2017, 09:01

why madvr outputs same picture no matter which matrix is reported by upstream?
To me they look different.
https://abload.de/img/bt709vpus6.png
https://abload.de/img/bt2020nc_f2uo2.png

Nintendo Maniac 64

10th June 2017, 01:52

If you mean software offline decoding (downloaded stream decoded locally from a HDD) then Core i5 is more than enough.

But online software decoding of 4K60fps VP9 using latest Chrome x64 is not feasible with Core i5.

In that case you should be just fine with a browser that uses the faster ffvp9 decoder that MPC-HC uses, such as Firefox and/or its forks (like Pale Moon, particularly the 64bit variants).

The vpx decoder that Chrome uses is known to me considerably slower by comparison (even with 64bit Chrome).

NikosD

10th June 2017, 05:29

I hate Firefox.

I'll wait for AMD to fix the drivers.

Thanks everyone.

mzso

10th June 2017, 09:26

I hate Firefox.

I'll wait for AMD to fix the drivers.

Thanks everyone.

Then don't waste other people's time by claiming/trolling "no other browser is suitable for that"!

NikosD

10th June 2017, 09:31

My initial question was regarding Chrome and hardware acceleration.

And I think it's true that Chrome has some privileged access to YouTube streams.

Anyway, if you are a simple fan or fanboy of Firefox, you are free to not reply to my posts, but you are certainly not allowed to speak for the others.

huhn

10th June 2017, 10:38

This is just not true all. I just looked at an UHD 60fps video and it plays in VP9. Just tried it in firefox stable for your sake and you only need to flip media.mediasource.webm.enabled to true in about:config, and then you get served vp9 videos just like in Chrome. So you can give it a try if it's better. (Also FF 54 is released in a week, which may, or may not perform better)

do you really need to flip media.mediasource.webm.enabled?
i didn't change anything ever and get VP9 for everything in like forever.

mzso

10th June 2017, 11:12

do you really need to flip media.mediasource.webm.enabled?
i didn't change anything ever and get VP9 for everything in like forever.

It looks like it. I wouldn't have thought so but when I tried with a new profile youtube only played mp4/avc files

i didn't change anything ever and get VP9 for everything in like forever.
It's not a new setting, so you might have enabled it a year or two ago, like I did. Even now on latest Nightly build it defaults to off.
I doesn't make sense to me, why implement something if you have it disabled for 99.9% of people

Anyway, if you are a simple fan or fanboy of Firefox, you are free to not reply to my posts, but you are certainly not allowed to speak for the others.
You don't get to decide. You sidetracked the conversation with your false, dishonest comment. You only had to say: "I'm a Chrome fanboy, so I don't care about better functioning other browsers."

huhn

10th June 2017, 13:05

It's not a new setting, so you might have enabled it a year or two ago, like I did. Even now on latest Nightly build it defaults to off.
I doesn't make sense to me, why implement something if you have it disabled for 99.9% of people

i can't deny that but i'm pretty sure i didn't change anything in the about:config ("this might void your warranty" i'm pretty sure i would remember this part). maybe it is enabled/disabled with different hardware/OS.

edit: well i guess that flip is not needed or i'm missing the topic right now: https://abload.de/img/strangeljsqy.png

mzso

10th June 2017, 13:34

i can't deny that but i'm pretty sure i didn't change anything in the about:config ("this might void your warranty" i'm pretty sure i would remember this part). maybe it is enabled/disabled with different hardware/OS.

edit: well i guess that flip is not needed or i'm missing the topic right now: https://abload.de/img/strangeljsqy.png

I don't know what are you doing but normally with that setting disabled firefox wont play or get vp9 videos on youtube.

huhn

10th June 2017, 13:47

mzso

10th June 2017, 14:15

what so ever that option doesn't matter for me.
but and that's important, hardware acceleration DOESN'T work with firefox. no issue for me but that's important for some systems. so if present does it work at your end?

well firefox and chrome have one thing in common a terrible downscaler and the software VP9 decoder should be the same ffmpeg decoder.

Maybe you somehow ran into a bug where that setting is bypassed. (BTW which FF version and OS are you using?)

I think Chrome just uses libvpx which is not as fast as ffvp9

I don't know what you mean by the downscaler. It looks fine to me. (It's actually better than EVR for downscaling which sucks on Nvidia/AMD GPU-s) (you see aliasing patterns)

huhn

10th June 2017, 14:48

firefox 53 stable 32 bit(why is the default installer not 64...)
windows 10 14393.1066

my HTPC is using an old windows 10 build and firefox 54 64 bit with the same result VP9 playback no hardware acceleration.
so i'm not sure if this about:config option matter what so ever.

in my personal experience firefox is unable to deal with high DPI setting and playback breaks down while playback generally works on chrome but far from perfect with doubling scaling because it can't deal with high DPI too.

the HTML5 downscaler is point or bilinear on my system the aliasing is out of this world and motion is a pain to watch with poping up pixels. well i have nothing good to say about web browser videoplayback anyway... i'm not disagreeing that EVR downscaling is bad but i can't remember it to be that bad.

mzso

10th June 2017, 14:52

firefox 53 stable 32 bit(why is the default installer not 64...)
windows 10 14393.1066

my HTPC is using an old windows 10 build and firefox 54 64 bit with the same result VP9 playback no hardware acceleration.
so i'm not sure if this about:config option matter what so ever.

Looks like there was an override added in bug 1230265 (https://bugzilla.mozilla.org/show_bug.cgi?id=1230265), mentioned here: https://bugzilla.mozilla.org/show_bug.cgi?id=1260305#c0

It seems like you just have fast computers.

the HTML5 downscaler is point or bilinear on my system the aliasing is out of this world and motion is a pain to watch with poping up pixels. well i have nothing good to say about web browser videoplayback anyway... i'm not disagreeing that EVR downscaling is bad but i can't remember it to be that bad.

I get no such things, on the other hand EVR looks horrible: https://forum.doom9.org/showthread.php?t=170711

huhn

10th June 2017, 15:07

NikosD

10th June 2017, 15:19

Hardware acceleration of VP9 using AMD Polaris cards is only working for YouTube using Chrome, officially.

mzso

10th June 2017, 15:31

leaves the mystery open why firefox is missing hardware acceleration for VP9.

not sure if a 4 year old i3 or a 5 year old i7 is consider fast but yes more than fast enough for VP9.

about the downscaling even this image show how bad it is: https://abload.de/img/strangeljsqy.png .it's a very low resolution no question could be the DXVA scaler but as far as i know HTML5 player are not using it.

Maybe it was just simply never implemented, and no-one cares...

As for downscaling, just verified it. It does suck. Though since I never play UHD videos and if I intend to watch a video I have it in full screen I don't see such artifacts.

Vincent Vega

10th June 2017, 15:43

To me they look different.
https://abload.de/img/bt709vpus6.png
https://abload.de/img/bt2020nc_f2uo2.png

thanks for that comparison! now i notice the difference taking screenshots like you did.
it's just barely noticeable (not like in the old days applying 601 or 709 to a dvd)

so bt2020nc is the correct one for those samples?
should i remux such "bt709" hdr samples with that setting?

also, i'm new to madvr, why in your screens i see:
chroma <
luma <
and my own for same video show:
chroma >
image <

any chance you could share your madvr settings?

i'm also still looking for quality settings that would allow me to use mpc-hc 'save image' correctly, while still being able to play 4k hfr hdr in realtime on gtx960 (and i7 6700k).

huhn

10th June 2017, 15:44

missing hardware acceleration is quite a good reason not to use firefox.

edge and chrome have it and firefox is automatic using the Superior VP9 version on capable systems so adding hardware decoding only makes sense. nvidia had a fixed function decoder for years now.

i totally understand why the hybrid decoder from polaris is not on by default. that would be a nightmare for 2 screen gamer...

Vincent Vega

10th June 2017, 16:35

do you really need to flip media.mediasource.webm.enabled?
i didn't change anything ever and get VP9 for everything in like forever.
It looks like it. I wouldn't have thought so but when I tried with a new profile youtube only played mp4/avc files

i just tried in ff 53, it seems that setting media.mediasource.webm.enabled to true forces it to skip its vp9 benchmark and enable it immediately.
if however you set it to default - false, and reset
media.benchmark.vp9.fps
media.benchmark.vp9.versioncheck
it won't be enabled at once, but after you access a page that offers vp9, it'll run the benchmark, create/populate the above two values, and if successful will begin to offer you vp9 upon the video page refresh.
example
https://www.youtube.com/watch?v=GVakZykycaI

huhn

10th June 2017, 16:46

"everything" is VP9 for me without a refresh. so there should be a general check for it and my firefox should know VP9 isn't a problem what so ever. how do you even see if it changes the settings or not?

mzso

10th June 2017, 16:47

@Vincent Vega
I think it's insane as hell. Made a setting then broke it on purpose. Then when one of them suggested a new setting to enable/disable probing to add some reason to this madness they wontfix-ed it...

Nintendo Maniac 64

11th June 2017, 02:01

mzso

11th June 2017, 11:38

Can I mention Pale Moon? It's downscaling from 4k looks fine to me and it supports VP9 + MSE if you enable it via Options -> Content.

I believe it also uses the ffvp9 decoder, so it should have good software decoding performance as well.
(Well, this is a VP9 thread, not a Firefox/Chrome thread. :) )

So Pale Moon caught up bit backend-wise? I've used it for a while maybe two years ago. But it stopped updating the firefox backend. So lacked proper html5 video support and such.

If downscaling's good, I wonder if firefox has a setting that can be enabled for decent downscaling. (I wouldn't be surprised)
Can you find out which about:config setting that option changes in PM?

Edit: I took a look, I see now that I misunderstood. You were talking enabling vp9.
Anyway, you don't seem to be right. If anything, it's worse but certainly not better:
https://abload.de/img/pm-downscaleajs3o.png

Nintendo Maniac 64

12th June 2017, 02:38

Anyway, you don't seem to be right. If anything, it's worse but certainly not better:
https://abload.de/img/pm-downscaleajs3o.png

Sorry, it's only high quality down to a specific size (I don't know the exact size, but I believe it's anywhere between 30% and 50%).

I did not realize you were trying to downscale to like 15% of the original size.

Now for images, there is actually an about:config setting that can be modified that makes it use the high-quality downscaling for all sizes (setting "image.high_quality_downscaling.min_factor" to '-1'), but AFAICT there isn't a similar setting for video (I will ask on the Pale Moon forum if this is the case).

nakTT

16th June 2017, 01:41

Hi guys.

Is there any way for us to actually filter/search YouTube content for VP9 encoded videos?

Thank you in advance.

mzso

16th June 2017, 07:05

Hi guys.

Is there any way for us to actually filter/search YouTube content for VP9 encoded videos?

Thank you in advance.

I doubt it.
I don't see the use anyway these days. Since the vast majority now has VP9 streams.

NikosD

20th July 2017, 12:35

During my little investigation for broken Chrome VP9 hybrid acceleration of Youtube using a Polaris card, I read about two things:

1) Latest Firefox v55 beta 8 has enabled VP9 hardware acceleration.

Could someone elaborate on that ?

2) Although there is hardware acceleration support of 8K VP9 on hardware like Kabylake or Pascal cards and MS supports that resolution via DXVA2 (and we have seen codecs like LAV filters actually leveraging that HW acceleration) there are exactly zero browsers that can take advantage of that and offer 8K VP9 HW acceleration.

I read that browsers support goes up to 4K VP9 for hardware acceleration.

I have a Polaris card and I can't check it out.

Could someone with a Pascal card or Kabylake check it out ?

Vincent Vega

20th July 2017, 18:42

i enabled this in ff54 a few weeks ago, hw vp9 decode now works (gtx 960)
media.wmf.vp9.enabled -> true

Official 20170611 builds are out (http://forums.mozillazine.org/viewtopic.php?p=14751697&sid=c34fffc1387d2735e78325bfd62c33e3#p14751697)

media.eme.vp9-in-mp4.enabled - also set that one to true.

NikosD

21st July 2017, 09:43

Well, I tried latest Firefox x64 v55 beta 10 with all possible options forced enabled and there is no difference in CPU utilization between VP9 HW acceleration and VP9 software decoding.

It immediately fetched the 4K60fps VP9 streams without "hacking" of the about:config options but with no visible acceleration.

I even set media.hardware-video-decoding.force-enabled to true, but with no luck.

In fact, trying to decode the Youtube stream of the 4K60fps clip here https://www.youtube.com/watch?v=ppy_gjBp71E, using Firefox is the worst experience compared with to the other browsers, because it turned out like a slideshow !

NikosD

21st July 2017, 11:59

VLD decoding can be implemented by pure or hybrid decoders.

Of course hybrid decoding is a lot more difficult and the work is done by the drivers - in our case AMD.

AMD had and has built MFT decoders for H.264 & H.265 and as a matter of fact H.264 HW MFT decoders were leveraged by Flash/H.264 Youtube streams by every browser out there.

There is no reason the same thing, can't be reproduced by AMD for VP9 hybrid VLD MFT decoder.

Vincent Vega

21st July 2017, 12:03

here it plays nice and smooth by gpu, up to 4k60 vp9 (that's max res GM206 supports i think)
https://developer.nvidia.com/nvidia-video-codec-sdk#NVDECFeatures

https://s18.postimg.org/c9xy92shl/vp9_4k60_hwdecode_gtx960_ff54.png

NikosD

21st July 2017, 12:06

here it plays nice and smooth by gpu, up to 4k60 vp9 (that's max res GM206 supports i think)

Perfect indeed.

Have you tried out your Skylake iGPU on the same stream VP9 4K60fps ?

Vincent Vega

21st July 2017, 12:19

NikosD

21st July 2017, 12:23

For me the best way is always by connecting two cables at the same monitor one e.g HDMI with dGPU and the other DVI/HDMI with iGPU.

Then it is extremely easy from the control panel to switch GPUs and from the monitor to switch signal source (port of the monitor)

I haven't found out a better way to have DXVA2 native acceleration by any GPU I want (dGPU or iGPU)

NikosD

22nd July 2017, 08:05

you mean in browser or media player?

in mpc-hc i use dxvacb, it just uses dGPU by default, somebody mentioned somewhere a hack to force iGPU but i dont remember where it was
in browser i dont know how to force use of iGPU, i could try if you link to instructions

Your Skylake 6700K would be the perfect test in order to check out what is going on with VP9 hybrid acceleration of Chrome/Firefox/Edge using Youtube, because Skylake like Polaris cards has only GPU accelerated (hybrid) VP9 decoding support.

So, if you manage to activate your iGPU, take a look on CPU and iGPU utilization during Youtube 4K60fps clip decoding.

It would be best to try all the browsers mentioned above (Chrome, Firefox, Edge)

UPDATE:
According to this patch to Chrome all Intel CPUs/GPUs with hybrid VP9 acceleration - prior to Kabylake that has fixed-function VP9 decoder - are disabled by default because they are too slow.

https://groups.google.com/a/chromium.org/forum/#!topic/chromium-reviews/BghP-1eqyhg

@Vincent Vega

You should probably force enable VP9 acceleration of Chrome for Youtube using your Skylake, like Polaris cards, but I doubt it will work due to luck of iGPU's performance.

Maybe Edge and Firefox allow it by default or else just use the following instructions for force enable VP9 acceleration on those browsers too.

Force enable VP9 acceleration:

For Chrome:

chrome.exe --ignore-gpu-blacklist --enable-accelerated-vpx-decode=2 for AMD

chrome.exe --ignore-gpu-blacklist --enable-accelerated-vpx-decode=3 for everyone

For Edge:

Write in the Edge browser address about:flags and then go to Media Source Extensions and set Enable VP9 video format option from Automatic to Always On

For Firefox:

Write in the Firefox browser address about:config and set media.mediasource.webm.enabled to true

I also set media.hardware-video-decoding.force-enabled to true

Vincent Vega

22nd July 2017, 19:22

hey man, sorry for late reply and thanks for the detailed overview!

i dont have chrome installed here, but, as to the previous dGPU decode talk, in Edge VP9 hw decode by gtx960 works just as fine as in firefox (with 'Enable VP9 video format' remaining at Automatic)

as to activating iGPU for the hybrid decode tests by monitor connection - is not possible right now for this here is more of a test platform build with just an old analog single-in monitor connecting to dGPU dvi-i. so unless there's some hack way to force browser to use the unconnected (and with digital-only outputs) skylake this method couldnt work. so need some trick setting to enable it in such config..
.

Nintendo Maniac 64

23rd July 2017, 01:16

i dont have chrome installed here

There's always the portable version:

http://portableapps.com/apps/internet/google_chrome_portable

NikosD

23rd July 2017, 03:21

As to activating iGPU for the hybrid decode tests by monitor connection - is not possible right now for this here is more of a test platform build with just an old analog single-in monitor connecting to dGPU dvi-i. so unless there's some hack way to force browser to use the unconnected (and with digital-only outputs) skylake this method couldnt work. so need some trick setting to enable it in such config.

I'm 100% sure that 950/960 and all Pascal cards as Kabylake, work immediately and out of the box as VP9 Youtube accelerators for all three browsers.

The interesting part of VP9 Youtube acceleration is with Broadwell & Skylake hybrid VP9 acceleration for Intel and Polaris & older GCN cards for AMD.

In your case with your setup the only thing I could think about is a DVI-TO-VGA adapter to connect directly your Skylake to the monitor or bring another monitor to your test system with digital input ports.

Vincent Vega

23rd July 2017, 17:43

skylake would require external active digital->analog signal converter, or similar onboard implementation on mobo, it natively outputs only digital, analog-out support was discontinued by intel starting SKL generation. probably just someone else with normal display and skylake could report these results here sooner for all of us..

what i was still curious about is maybe finding a way to use skl hw decode without connected monitor.
like for example, with this very setup in LAV decoder/media player i can choose quicksync decode, e.g. for h264, and it will do it by iGPU, and yet output to monitor connected to dGPU.
though LAV setting says i cannot do the same for dxva2cb/dxva2n, only dGPU available with these for me, because it connects to monitor.
so i dunno whether hw/hybrid decoding in browser is subject to same limitation or not.

NikosD

25th July 2017, 23:03

Good news!
It seems that latest Chrome Canary v62 (x64) works like a charm using VP9 hybrid acceleration of a Polaris card and Creators Update!

Firstly, like plain Chrome v59 (x64) it recognizes out of the box VP9 Youtube content but it doesn't enable by default the hybrid VP9 acceleration.

BUT if you force enable VP9 acceleration using the --enable-accelerated-vpx-decode=2 switch, you see no stuttering in the video decoding and it has a minimal frame dropping of just 35 frames during the whole duration of the very difficult 4K60 fps Youtube clip posted above.

The GPU load works exactly like Chrome, but without stuttering and you see the GPU and memory clocks going to the highest levels.
The CPU usage is around ~50% on average.

So, it seems that AMD's VP9 hybrid acceleration works, at least on one Chromium based browser and force enabled.

leonccyiu

24th September 2017, 02:50

So it appears that the Apple TV 4k won't play back 4k or hdr videos on Youtube because Apple won't support VP9 and Google doesn't encode 4k in h264 anymore.

This will affect future iOS devices as well such as the iPad Pros, and the unlikely event that a future iPhone has a beyond 1440p resolution.

Instead of taking a side, I am going to say that both these companies are despicable using their leverage and market positions as monopolies to force each others hands.

I understand that VP9 is open and royalty free, but Google owns patents to the format and requires those who adopt it to give up existing patent disputes. Also Google is the one who stopped encoding 4k in h264.

Apple on the other hand are always using proprietary formats such as HEIF for photos in iOS 11 to try and lock users into their ecosystem.

The only losers are ordinary consumers and not these multi billion dollar companies.

If Google didn't have a monopoly on user uploaded streaming video and Apple wasn't such a big player in mobile devices (I have an android phone and want to switch to iOS due to stability issues), such a format war and dispute would be much less likely to happen.

x265_Project

24th September 2017, 04:17

...Apple on the other hand are always using proprietary formats such as HEIF for photos in iOS 11 to try and lock users into their ecosystem.

Standards like HEIF and HEVC are not owned by Apple, they are owned by all of the companies that contributed to the standard. Apple can't lock users in to their ecosystem when they adopt a worldwide standard that is also adopted by many hundreds of companies. Samsung, LG, and Sony also produce premium smartphones that support HEVC, and we hope they also choose to support HEIF, which is 2.4x more efficient than JPEG.

Both companies take a risk by not supporting the format chosen by the other company. YouTube takes a risk that their videos don't look as good as they should on many devices (4K TVs, Apple TV, etc.), and Apple takes a risk that consumers will choose another connected set-top that supports 4K YouTube videos. When one of them starts losing market share to a competitor that points out their shortcomings, things will change. Only time will tell how this all plays out.

hajj_3

24th September 2017, 10:02

I understand that VP9 is open and royalty free, but Google owns patents to the format and requires those who adopt it to give up existing patent disputes. Also Google is the one who stopped encoding 4k in h264.

Firstly i'm pretty sure they are only required to give up existing vp8/vp9 patent disputes.

Secondly, you can't complain that google doesn't want to encode 4k videos in h264, that require a huge amount of bandwidth and thus is expensive. Are you willing to donate google $50m or so to pay for the videos to use h264 instead? I thought not.

Apple are the one being the idiots, they should just add vp9 to safari and to their apple tv box.

mzso

24th September 2017, 15:26

Apple are the one being the idiots, they should just add vp9 to safari and to their apple tv box.

I wonder what they'll be doing when AV1 comes out...

IgorC

24th September 2017, 19:37

and we hope they also choose to support HEIF, which is 2.4x more efficient than JPEG.

It's not a first time when You make such claims without actually providing any proof.

So provide a link or your post would be considered as it should be (as a lie) or at least misinformation, marketing, etc...

AFAIK there was no one single study that confirm HEVC/HEIF is 2.4x more efficient (subjectively, not PSNR!) than JPEG.

x265_Project

24th September 2017, 20:11

See Table IV... https://nokiatech.github.io/heif/technical.html and http://phenix.int-evry.fr/jct/doc_end_user/documents/22_Geneva/wg11/JCTVC-V0072-v1.zip

On average, JPEG requires a bit rate that is 139% higher (2.39x) than HEVC to achieve identical quality to HEVC Intra compression. I rounded this to 2.4x.

IgorC

24th September 2017, 20:33

The study You mention is based on PSNR results and obscure encoders and techniques.

http://phenix.it-sudparis.eu/jct/doc_end_user/current_document.php?id=7292
Page 2:

The PSNR of the resulting pictures are recorded and pictures are coded so that the PSNR’s match for other codecs.

It's PSNR! Do not tell that You didn't know. Just don't, ok?

So, you are perfectly fine with "comparisons" where HEVC/HEIF is superior even if it's PSNR or even worse if there were made by the same patents holders like Fraunhofer https://forum.doom9.org/showthread.php?t=174540

But You are extremely against PSNR in your own blog
http://x265.org/compare-video-encoders/

Since this moment your statements are worthless.

P.S. Here is more realistic which includes subjective comparison (not just metrics)
https://jpeg.org/downloads/aic/wg1n73041_icip_2016_grand_challenge.pdf
HEVC and Daala are 1.4-1.65x more efficient than JPEG according to this study

It's hardly possibility than even AV1 (not sure about H.266) would have 2x JPEG's efficiency for image compression.

x265_Project

24th September 2017, 23:27

Yes, of course I agree that subjective evaluation is the only way to truly compare 2 codecs. The study I linked to above says...

"Subjective testing performed for different test sets appear to verify the results also when it comes to the perceived quality of material coded with different codecs. For instance, [6] reports that typically subjective quality of HEVC intra coded pictures are comparable to that of JPEG coded pictures using twice, or sometimes even four times the bitrate of HEVC."

http://phenix.int-evry.fr/jct/doc_end_user/documents/12_Geneva/wg11/JCTVC-L0380-v2.zip

HEVC and Daala are 1.4-1.65x more efficient than JPEG according to this study
I read the presentation you linked to, but I don't see where you got your number from. The subjective evaluation scores for HEVC were clearly much, much higher than for JPEG, but there didn't appear to be enough data points to really measure the difference in compression efficiency.

When a device has HEVC encode/decode capability (which the leading premium smartphones do), it obviously makes sense to use that for photo compression instead of JPEG. If you're shooting in burst mode, you can get another 2 to 5x more compression efficiency from HEIF, which can encode the image sequence as a video, using inter-prediction (P frames and B frames).

IgorC

25th September 2017, 00:40

http://phenix.int-evry.fr/jct/doc_end_user/documents/12_Geneva/wg11/JCTVC-L0380-v2.zip

It says:
"The evaluation results demonstrate that HEVC intra coding outperforms encoders for still images with an average bit rate reduction ranging from 16% (compared to JPEG 2000 4:4:4) up to 43% (compared to JPEG)."

So where is your 2.4x efficiency over JPEG?

x265_Project

25th September 2017, 20:00

It says:
"The evaluation results demonstrate that HEVC intra coding outperforms encoders for still images with an average bit rate reduction ranging from 16% (compared to JPEG 2000 4:4:4) up to 43% (compared to JPEG)."

So where is your 2.4x efficiency over JPEG?

Igor - Did you look at the RD curves in the subjective test you linked to? Every example shows HEVC achieving better subjective quality at half or less than half of the bit rate of JPEG.
bike.ppm
0.25 bpp HEVC MOS = 2.75
0.5 bpp JPEG MOS = 2.0
HEVC looks measurably better at 1/2 the bit rate

cafe.ppm
0.5 bpp HEVC MOS = 4.3
1.5 bpp JPEG MOS = 3.5
HEVC looks significantly better at 1/3 the bit rate

The same is true for every other photo shown - at least 2x better efficiency is demonstrated (especially at very low bit rates). Of course, the % improvement in compression efficiency varies depending on the image, the chroma subsampling and the quality levels being compared. Give any encoder enough bits and it will be lossless.

Mozilla had a nice study posted a couple of years ago, but the links don't work anymore for some reason. I think most people would be thrilled to increase their photo storage capacity by another 57% (the storage increase you get when you reduce file size by 43%), but I believe we'll see that overall, the increase in efficiency is higher than 43%.

Edit - we've veered off topic. We should probably be having this discussion on the HEVC forum, not a VP9 thread. Mods... feel free to move these posts.

IgorC

26th September 2017, 00:18

Cherrypicking at its best. :rolleyes:

I can cherrypick the same way.

https://jpeg.org/downloads/aic/wg1n73041_icip_2016_grand_challenge.pdf

p08.ppm
JPEG is slightly better than HEVC at 1 bpp

woman.ppm
JPEG is slightly better than HEVC at 0.75-1 bpp

So not only HEVC isn't 2.4x efficient (as You claim) there are images where is worse than JPEG. Look the grpah for yourself.

Calculating an average of all 6 subjective results it can be said that Daala/HEVC were approx ~1.6-1.7x of JPEG efficiency at MOS 4(perceptible but not annoying). And the efficiency is 1.2-1.4x at MOS 4.5

x265_Project, you aren't an average forum member. This is serious stuff. You're developer and You've lied about 2.4x efficiency of HEVC over JPEG and lately have tried cherrypicking on results. :)

x265_Project

26th September 2017, 01:59

x265_Project, you aren't an average forum member. This is serious stuff. You're developer and You've lied about 2.4x efficiency of HEVC over JPEG and lately have tried cherrypicking on results. :)
You're free to disagree with me, but it's another thing to call me a lier, even if you tried to soften the blow with a smiley face. I haven't lied about anything. If you want people to respond to you on a public forum like this, you need to stick to the subject (the technology), and avoid personal accusations or insults.

It's not "cherrypicking" to point out that there is a place along the RD curve where the efficiency gain of one codec is 2x or greater than another codec. That's the way RD curves work. That's why I pointed out that " the % improvement in compression efficiency varies depending on ... the quality levels being compared." Pick a different place along the 2 curves (as you tried to), and you'll see a different gain.

It's not possible quantify the efficiency gain of Codec A over Codec B in with perfect precision and get universal agreement, because the result always depends on the content, the settings (like Chroma subsampling), and the PLACE ON THE RD CURVEs (the quality level) you choose in your test. Change any variable, and you'll get a different result (especially the place on the RD curve that you're attempting to compare).

The only place along the RD curve that no one can debate is the highest quality - lossless compression. Certainly the % bit rate savings will be different for lossless quality than for typical lossy compression settings. But that doesn't mean that it isn't valid to say that "for typical lossy compression bit rates the savings is X". We could argue ad infinitum as to what bit rates are typical (what point on the RD curve we should compare). But that would be a pissing contest.

We're discussing/debating a very limited set of data from a couple of studies that are several years old, using small picture sizes (800x1280). To add more clarity and arrive at a consensus we need more data, using a wider range of content, picture sizes and quality levels, as well as the latest builds of HEVC encoders (x265, HM reference encoder, etc.).