Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion. Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules. |
12th May 2014, 14:44 | #1 | Link | |
Registered User
Join Date: Jul 2011
Posts: 1,121
|
YouTube now adding VP9 option to popular videos
Will that be decodable with the default VP9, or will the branch have to merge or something?
Quote:
Fun fact though, Youtube have started to use it on Videos as a "Standard", it seems that they encode all popular videos (about 10k views) to VP9 according to my fast research. What i would like to know though is. How in the world are they able to encode all the videos if VP9 is that slow, even if they only do 10k videos, it's still pretty insane. Never the less, it's not all 10k videos, just overall, also the bitrate compared to mp4(h264) is less. 1080p is about 3-4k bitrate with h264 if i remember correctly, VP9 is about 2.5, taken the numbers from my head right now so take it with a grain of salt. I have also compared the encoding of some videos, and well, they beat each other at different things. Overall however, h264 wins, which i find quite surprising as the settings they use for it it's extremely bad. |
|
12th May 2014, 23:27 | #3 | Link |
Registered User
Join Date: Nov 2009
Location: Northeast Ohio
Posts: 447
|
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 [2015-12-20] Lossless 1080p, 2x speed @ 60fps [2015-11-06] Lossless 1080p, 30fps [2015-11-06] Lossless 1080p, 2x speed @ 60fps [2015-07-05] Lossless 1080p, 30fps [2015-05-01] 6Mbps AVC 720p, 30fps [2014-05-12] 6Mbps AVC 720p, 30fps [2013-11-12] 3/6Mbps AVC 720p, Half speed @ 15fps 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 ~ Screenshot B ~ Screenshot C 2014 DASH AVC (2112kbps): Screenshot A ~ Screenshot B ~ Screenshot C fmt22 AVC (2886kbps): Screenshot A ~ Screenshot B ~ Screenshot C WebM VP9 (2182kbps): Screenshot A ~ Screenshot B ~ Screenshot C 2015 DASH AVC (2125kbps): Screenshot A ~ Screenshot B ~ Screenshot C fmt22 AVC (2894kbps): Screenshot A ~ Screenshot B ~ Screenshot C WebM VP9 (1686kbps): Screenshot A ~ Screenshot B ~ Screenshot C -------------------------------- 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/interne...hrome_portable Firefox ESR: http://portableapps.com/apps/interne...x-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/fir...and-audio-dow/ Last edited by Nintendo Maniac 64; 25th October 2016 at 05:39. |
12th May 2014, 23:45 | #4 | Link | |
Registered User
Join Date: Oct 2009
Posts: 930
|
Quote:
Also vp9 has ~50% more bandwidth here. Which is a LOT. Last edited by mzso; 12th May 2014 at 23:52. |
|
13th May 2014, 00:05 | #5 | Link |
Registered User
Join Date: Nov 2009
Location: Northeast Ohio
Posts: 447
|
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... Last edited by Nintendo Maniac 64; 13th May 2014 at 01:49. |
13th May 2014, 02:39 | #6 | Link |
Registered User
Join Date: Dec 2012
Posts: 197
|
Couple of still shots from the 720p DASH formats.
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...). Last edited by xooyoozoo; 13th May 2014 at 02:56. |
13th May 2014, 04:01 | #7 | Link | |
Registered User
Join Date: Nov 2009
Location: Northeast Ohio
Posts: 447
|
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: Quote:
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/fir...and-audio-dow/ Last edited by Nintendo Maniac 64; 9th December 2014 at 09:03. |
|
13th May 2014, 11:31 | #9 | Link | |
Registered User
Join Date: Oct 2009
Posts: 930
|
Quote:
|
|
13th May 2014, 17:02 | #10 | Link |
Registered User
Join Date: Jul 2011
Posts: 1,121
|
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. Last edited by zerowalker; 13th May 2014 at 17:13. |
13th May 2014, 20:49 | #11 | Link | ||
Registered User
Join Date: Nov 2009
Location: Northeast Ohio
Posts: 447
|
Quote:
Quote:
Therefore pretty much anything 1080p is completely out of the question for me. |
||
13th May 2014, 20:55 | #12 | Link | |
Registered User
Join Date: Jul 2011
Posts: 1,121
|
Quote:
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. Last edited by zerowalker; 13th May 2014 at 21:12. |
|
13th May 2014, 21:16 | #13 | Link | ||||
Registered User
Join Date: Nov 2009
Location: Northeast Ohio
Posts: 447
|
Quote:
Quote:
Quote:
(that enough information for you? :P) ----------------------------------------------------------- Quote:
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. 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) |
||||
13th May 2014, 21:34 | #14 | Link | |||||
Registered User
Join Date: Jul 2011
Posts: 1,121
|
Quote:
Quote:
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? Quote:
Quote:
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 Quote:
Last edited by Guest; 13th May 2014 at 21:38. Reason: 3 |
|||||
13th May 2014, 21:50 | #15 | Link | ||
Registered User
Join Date: Nov 2009
Location: Northeast Ohio
Posts: 447
|
Quote:
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) Quote:
(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) I'm seeing 50-80% on just 720p 30fps VP9, I haven't even tried 1080p yet. Last edited by Nintendo Maniac 64; 13th May 2014 at 21:53. |
||
13th May 2014, 22:09 | #16 | Link | ||
Registered User
Join Date: Jul 2011
Posts: 1,121
|
Quote:
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 Quote:
Last edited by Guest; 13th May 2014 at 22:11. Reason: 3 |
||
13th May 2014, 22:14 | #17 | Link |
Registered User
Join Date: Nov 2009
Location: Northeast Ohio
Posts: 447
|
I've already downloaded four 720p VP9 videos that are all like 30 minutes. 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. Last edited by Nintendo Maniac 64; 18th June 2015 at 08:58. |
14th May 2014, 11:11 | #18 | Link | |
Registered User
Join Date: Jun 2002
Posts: 35
|
Quote:
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/tr...er/DOCS/man/en |
|
14th May 2014, 11:42 | #19 | Link | ||
Registered User
Join Date: Jan 2007
Posts: 729
|
Quote:
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. Quote:
Last edited by mandarinka; 14th May 2014 at 11:46. |
||
|
|