x265 HEVC Encoder

[benwaggoner]

Where are people at with using --fades in x265 3.5? Anyone see it make a difference, positive or negative?

From the readthedocs documenation:
--fades, --no-fades
Detect and handle fade-in regions. Default disabled.

From the 3.1 release notes:
--fades can detect and handle fade-in regions. This option will force I-slice and initialize RC history for the brightest frame after fade-in.

[charliebaby]

I did a test with the --fades but I don't see any difference

[FranceBB]

I did a test with the --fades but I don't see any difference

Check the GOP Structure with:

ffprobe.exe -show_frames -i "Test.ts" | findstr pict_type

pause

It will show you each frame and which kind of frame it is, so basically I, P or B.
Technically each scene change should be an I.
Encode the same content twice, same parameters, one with --fades the other without and compare the GOP and see if it changes. If it does, then it's working.

[Blue_MiSfit]

Are there any actual changes other than commit value in this build? It appears all 2022 commits (https://bitbucket.org/multicoreware/x265_git/commits/) have been for ARM (courtesy of AWS).

Very cool, thanks Amazon :)

I've been meaning to build an ARM ffmpeg docker image and give the latest Graviton instances a whirl!

[quietvoid]

H.274 film grain characteristics SEI support in x265: https://mailman.videolan.org/pipermail/x265-devel/2022-March/013459.html
Also made available, a tool to generate the metadata from YUV: https://bitbucket.org/multicoreware/libfgm

Have fun!

[Boulder]

Very interesting! Does it require specific support from the decoder or is it a standard requirement in the specification?

[quietvoid]

I'm pretty sure it does require support at the decoder/renderer. I don't know whether current hardware supports it.
FFmpeg can decode and apply one of the types at 8 bits only as far as I can tell.

[nevcairiel]

Very interesting! Does it require specific support from the decoder or is it a standard requirement in the specification?

It is not a standard requirement for HEVC, it is optional SEI data.
AV1 has made it a baseline function, for what its worth, including support in HW decoders.

[charliebaby]

New Version x265-3.5.0.36-3415763_gcc112-AVX2

GCC 11.20

https://www.mediafire.com/file/ssl3me9ghx6s4x7/x265-3.5%252B36-3415763_gcc112-AVX2.exe/file

[Barough]

x265 v3.5+37-9af013bb3
Built on March 21, 2022, GCC 11.2.0

https://www.mediafire.com/file/hokjj9zy2afu0p0/x265-3.5%252B37-9af013bb3_Win_GCC112.7z/file


Note :

The commit value is wrong due to something upstream at MulticoreWare.

Correct commit number for this release is

9b59d45

[benwaggoner]

It is not a standard requirement for HEVC, it is optional SEI data.
AV1 has made it a baseline function, for what its worth, including support in HW decoders.
...although one of the most popular first generation AV1-capable SoCs had a bug with FGS which makes it problematic to safely use. I think it can be fixed with a driver update, but lots of devices don't regularly update those sorts of components.

[LazyNcoder]

Guys, using aq-mode 4 with CRF, I'm getting much smaller file. Is this OK?
Obviously we're losing some details here. Should I choose a lower number for CRF to keep the same quality? Is it enough? maybe some more aq-strength? little bite psy-rd and psy-rdoq?

[rwill]

Guys, using aq-mode 4 with CRF, I'm getting much smaller file. Is this OK?
Obviously we're losing some details here. Should I choose a lower number for CRF to keep the same quality? Is it enough? maybe some more aq-strength? little bite psy-rd and psy-rdoq?

I don't know if its OK, I don't know how it looks.

[benwaggoner]

Guys, using aq-mode 4 with CRF, I'm getting much smaller file. Is this OK?
Obviously we're losing some details here. Should I choose a lower number for CRF to keep the same quality? Is it enough? maybe some more aq-strength? little bite psy-rd and psy-rdoq?
It's okay if it's giving you better bang for the bit, which is hard to figure out when both bitrate and detail drops. The only real way to test is to compare two different 2-pass VBR encodes using the same average bitrate but different AQ modes.

[DKILLER1]

My anime encoding settings is (720p)

--ctu 32 --max-tu-size 16 --tu-inter-depth 3 --splitrd-skip --tu-intra-depth 3 --b-intra --ssim-rd --weightb --cbqpoffs -3 --crqpoffs -3 --no-open-gop --max-merge 4 --rc-lookahead 90 --lookahead-slices 0 --limit-refs 1 --limit-modes --rdpenalty 1 --ref 5 --me 3 --psy-rd 2.40 --subme 4 --bframes 6 --merange 50 --psy-rdoq 3 --rdoq-level 2 --deblock -3 --aq-mode 2 --aq-strength 1.30 --rd 4 --no-strong-intra-smoothing --no-sao --limit-tu 2 --qg-size 16 --qcomp 0.70 --fades

Is there anything I should change to further improve the quality of the anime?
Please anyone help here

[benwaggoner]

My anime encoding settings is (720p)
Is there anything I should change to further improve the quality of the anime?
Please anyone help here
How did you wind up with these settings? A lot seems pretty random. For example, --rdpenalty only applies to 64x64 CUs, but you're using --ctu 32. And I've never seen --max-tu-size 16 used for anything. Did you actually see improvements by setting that?
Or --ssim-rd, which didn't work well for general content, and I would think would be less applicable to anime.

You're not specifying --profile or --level-idc, which can be important for compatibility.

I'd start with --preset slower --tune animation and iterate from there. It's rare that you'd need or should set so many parameters manually.

[LigH]

New upload: x265 3.5+37-0d865ebe2 (https://www.mediafire.com/file/znuc3jsmtsnx7jc/x265_3.5+37-0d865ebe2.7z/file)

[Windows][GCC 12.1.0][32/32XP/64 bit] 8bit+10bit+12bit

Bugs not yet fixed:

NASM 2.15.05 multi-line macro warnings (reported August 2020)

[LoRd_MuldeR]

New upload: x265 3.5+37-0d865ebe2 (https://www.mediafire.com/file/znuc3jsmtsnx7jc/x265_3.5+37-0d865ebe2.7z/file)

[Windows][GCC 12.1.0][32/32XP/64 bit] 8bit+10bit+12bit

Bugs not yet fixed:

NASM 2.15.05 multi-line macro warnings (reported August 2020)


:thanks:

[Jamaika]

https://github.com/netwide-assembler/nasm
How can I compile or download nasm 2.16?
There are supposedly newer nasm replacements. Where and from where can I download?

[LigH]

Looks like the right place. Pull it with git, adding .git to the URL. Or download an archive from the green button [ Code ] - "Download ZIP". Then ./configure, make ... as usual. I guess.

I should possibly suggest to the MSYS2 team that nasm 2.16 is available.

Or not. This tag is not "released" yet. The "latest" snapshot is 2.16rc0-20201104 ... hmm, it aged well.

[LoRd_MuldeR]

https://github.com/netwide-assembler/nasm
How can I compile or download nasm 2.16?
There are supposedly newer nasm replacements. Where and from where can I download?

What about?
https://www.nasm.us/pub/nasm/snapshots/latest/win64/

[Jamaika]

Sorry snapshots has bugs and it isn't 2022.

[tuanden0]

I am building my new PC for x265 video encoding because my PC is broken by thunderstruck accident.

Could everyone give me some advice?

My old CPU is R7 5800x with encode --preset veryslow about 6.2 fps (x265 version 3.5), I want to increase the encode speed at least 14 fps

[Khun_Doug]

My build was using a 3950X but I upgraded to a 5950X, 16 cores / 32 threads. The prices have dropped as availability increased. I encode with grain and slow selected, usually at CRF 20. That 5950X comes to life and moves along quite well. One factor that effects speed is the number of filters you have applied. I almost always encode with AVS, and set the AVS threads to 14. My experience was that anything higher than 14 on HD source didn't yield much benefit. On SD source I use a thread count of 6.

[rwill]

I am building my new PC for x265 video encoding because my PC is broken by thunderstruck accident.

Could everyone give me some advice?

My old CPU is R7 5800x with encode --preset veryslow about 6.2 fps (x265 version 3.5), I want to increase the encode speed at least 14 fps

Look at this thread, although some results are quite old:

https://forum.doom9.org/showthread.php?t=174393

So, just go on Lenovos website and start to look at Threadripper offerings I guess.

[RanmaCanada]

I am building my new PC for x265 video encoding because my PC is broken by thunderstruck accident.

Could everyone give me some advice?

My old CPU is R7 5800x with encode --preset veryslow about 6.2 fps (x265 version 3.5), I want to increase the encode speed at least 14 fps

On very slow nothing at a "reasonable" price will give you that. I would honestly say for now to get an inexpensive unit (sub $200 USD refurb) to get you buy until Zen 4 is released, as Raptor Lake will also be released "soon". It's possible there may be some insane increases in encode speeds (15% IPC at least rumoured). Even if there is not, Zen3 prices should drop some more and Intel 12th gen should as well as they have learned they can no longer demand stupid prices with AMD beating them.

[tuanden0]

Thank @Khun_Doug, @rwill, @RanmaCanada

I will follow RanmaCanada's idea to wait for AMD or Intel release new gen CPU to check perf/price to buy or waiting old gen price down price

[Jamaika]

I am going to ask this brazen question. Is this project finished?

[rwill]

I am going to ask this brazen question. Is this project finished?

It might be done for.

[RanmaCanada]

It might be done for.

Well hasn't it been in maintenance mode for quite some time now with no major updates? it's possible we might need to move to VVC or AV1 and completely relearn how to encode haha.

[filler56789]

I am going to ask this brazen question. Is this project finished?

We may say it is effectively-dead. :-|

* 2013
† 2022

Time flies indeed.

[Blue_MiSfit]

My understanding is that lots of talent left MulticoreWare, so the pace of development has slowed quite a lot.

[FranceBB]

Well I mean x265 is very mature now, so what I'm concerned about is actually x266.
It's been a while now since the official specs have been finalized and there are encoders like VVEnc from Fraunhofer which are somewhat usable compared to the original reference encoder which was meant to be used for tests only.
Decoders are also slowly but surely popping up and there are plugins that can be compiled and integrated in VLC etc.

And yet... no public x266 repository.
Given that there hasn't been any progress on x265 for a while, I assume Multicoreware is going hands down on x266 and focusing on it, but at this point I'm not sure.

H.266 VVC specs were finalized on the 6 July 2020, so in 2 days it's gonna be two years of no progress with x265 and still no x266. :(

[excellentswordfight]

Well I mean x265 is very mature now, so what I'm concerned about is actually x266.
It's been a while now since the official specs have been finalized and there are encoders like VVEnc from Fraunhofer which are somewhat usable compared to the original reference encoder which was meant to be used for tests only.
Decoders are also slowly but surely popping up and there are plugins that can be compiled and integrated in VLC etc.

And yet... no public x266 repository.
Given that there hasn't been any progress on x265 for a while, I assume Multicoreware is going hands down on x266 and focusing on it, but at this point I'm not sure.

H.266 VVC specs were finalized on the 6 July 2020, so in 2 days it's gonna be two years of no progress with x265 and still no x266. :(
I just had a look at MulticoreWares website and it seems a lot more focused on AI/ML than video encoding, same goes for their job openings.

But looks like they had a x266 demo at IBC this year though.

We may say it is effectively-dead. :-|

* 2013
† 2022

Time flies indeed.
Tbh it has pretty much been dead since 3.0 in 2019.

[FranceBB]

But looks like they had a x266 demo at IBC this year though.

IBC? I think you meant NAB. It can't be IBC 'cause it's in September and I'm gonna be there in person this year, hence my post: https://forum.doom9.org/showthread.php?t=184144

(which by the way is still valid as I'm gonna meet with steipal, momocampo and emcodem, but I'd love to met with more people from Doom9).

But anyway I'd love to see x266 on display there by the Multicoreware guys 'cause I'm gonna ask them a few questions and report everything here.
One of them is also gonna be whether they're gonna create an account called x266_Project like they did for the x265_Project from which they used to post everything regarding the development etc.

[Barough]

Double post

[Barough]

My understanding is that lots of talent left MulticoreWare, so the pace of development has slowed quite a lot.


After Tom left MulticoreWare so did it start to go downhill with the development. He was the Head of the x265 project and the Head of MulticoreWare's video business. He left back in the beginning of 2018.

[excellentswordfight]

IBC? I think you meant NAB. It can't be IBC 'cause it's in September and I'm gonna be there in person this year, hence my post: https://forum.doom9.org/showthread.php?t=184144

(which by the way is still valid as I'm gonna meet with steipal, momocampo and emcodem, but I'd love to met with more people from Doom9).

Yes, typo, i ofc meant NAB.

Are they confirmed for this years IBC as well? Not sure if I'm going this year, but if I do I guess I will have look as well.

[ksec]

After Tom left MulticoreWare so did it start to go downhill with the development. He was the Head of the x265 project and the Head of MulticoreWare's video business. He left back in the beginning of 2018.

He went to Beamr, which is arguably the best HEVC encoder. And now he is at Kaleidescape.

He doesn't tweet about Video Codec anymore. I have a feeling that he simply got fed up with the whole VVC licensing situation and simple left Video Codec field.

[rwill]

Well Video Compression is not a good business.

No one wants to pay more than a couple bucks for some encoder license, I mean there are ones free available.
If the free encoders suck people will complain but no one wants to sponsor development.
No one wants to pay for decoders, they already paid for the content. Some dont even pay for the content.
No one wants to buy extra hardware for encoding, why is a quad core CPU from 2012 not sufficient again ?
Some People with no clue mistake video compression for modern magic and then think they have three wishes free.
People then have elevated expectations about video quality, cant be hard to do better right ?
People don't want to spent time to learn about quality tunes, rather they want something like WinZip.

and I could continue this list ...

I mean I had the case where someone complained about quality and then someone asked him if he could try a higher bitrate and this person then was really thankful because a higher bitrate apparently improved quality a lot in his case. Quite the trick once you know it....

And the people you have to work with...

There are a lot of Pretenders that just have mastered the art of Bullshit Bingo.
The most capable people tend to be legally insane.


Does not mean that other areas of entrepreneurship do not have similar issues, but because software is immaterial and a video encoder or decoder is a non interactive black-box their value is hard to grasp for the average populance.

[benwaggoner]

I just had a look at MulticoreWares website and it seems a lot more focused on AI/ML than video encoding, same goes for their job openings.
That's always been a big focus of MCW, along with performance engineering in general. Codecs were never more than a fraction of their total business.

[Jamaika]

I mean I had the case where someone complained about quality and then someone asked him if he could try a higher bitrate and this person then was really thankful because a higher bitrate apparently improved quality a lot in his case. Quite the trick once you know it....
As they say in third world countries. Who uses low bitrate? Software pirate that used xvid interlaced DVD.
He doesn't know much about filming, framing, editing but he tries to sell something under the table.
There is still a 4 EURO DVD, not necessarily with voiceover. Then came x264 and x265. For years it wasn't as successful as xvid. Market collusion?
Initially x265 was better for low bitrates than Mainconcept HEVC in low-cost editors. Everyone disliked the fact that the copy of the original isn't webcam quality.
Youtube is now censored. Fired journalists pretend to contribute something in 4K. It no longer needs xvid, x264, x265.
Who needs Bluray for 16EURO today in Europe. There is no video DVD rental. We have pay TV for any high bitrate programs, websites, metadata as ECO

[benwaggoner]

IBC? I think you meant NAB. It can't be IBC 'cause it's in September and I'm gonna be there in person this year, hence my post: https://forum.doom9.org/showthread.php?t=184144

(which by the way is still valid as I'm gonna meet with steipal, momocampo and emcodem, but I'd love to met with more people from Doom9).
I'm planning on going myself! Let's have a meetup.

[benwaggoner]

Well Video Compression is not a good business.

No one wants to pay more than a couple bucks for some encoder license, I mean there are ones free available.
If the free encoders suck people will complain but no one wants to sponsor development.
No one wants to pay for decoders, they already paid for the content. Some dont even pay for the content.
No one wants to buy extra hardware for encoding, why is a quad core CPU from 2012 not sufficient again ?
Some People with no clue mistake video compression for modern magic and then think they have three wishes free.
People then have elevated expectations about video quality, cant be hard to do better right ?
People don't want to spent time to learn about quality tunes, rather they want something like WinZip.
If you are talking about a consumer market, yes, there's not a lot of money to be made in compression. The general class of workstation encoding tools largely vanished around a decade ago. To my regret; there are all sorts of high-touch operations that were easier to do back then which are much more challenging to do now. I barely did any command-line encoding the first half of my career, and now I barely do anything but.

However, enterprise and broadcast encoding is still a huge markets. At scale, bandwidth becomes quite expensive. And for RF-limited transmission like cable/sat/OTA, better compression=more channels=more revenue. A big cable company can spend $100M upgrading encoders with a healthy ROI.

Consumers and hobbyists largely have to make due to with plugging in scraps of open-source code funded by bigger corps these days. Advanced preprocessing is much more an enthusiast domain, but codecs themselves have gotten too complex for small ad-hoc teams to do the kinds of intense innovation like x265 had in its early days.

[FranceBB]

Who uses low bitrate? Software pirate

Not really, given how crowded satellites like HotBird are and that many many many TVs all over the globe rely on them to broadcast their contents, it makes sense to keep the development of new codecs ongoing.
And remember, it's those TVs that in the end are gonna pay the big bucks to get their fair share of bitrate and with ever growing resolutions, bit depth and frame-rates, encoders just need to get better.

Then you have a totally separate market which is VOD with streaming services which are also trying to get the best possible compression at the lowest possible bitrate 'cause CNDs ain't cheap either. I often say that satellite bitrate is hugely expensive, but think about companies like Amazon, Netflix etc and how many users they have and how much bandwidth they "spend" with their CDNs to serve all the customers.

I'm planning on going myself! Let's have a meetup.

Totally! :D
I'll PM my number just in case and I'll come up with more detailed info as the date approaches. :)


enterprise and broadcast encoding is still a huge markets. At scale, bandwidth becomes quite expensive. And for RF-limited transmission like cable/sat/OTA, better compression=more channels=more revenue. A big cable company can spend $100M upgrading encoders with a healthy ROI.

Absolutely.
I work one step ahead of the final distribution guys, so the files I encode are all high bitrate mezzanine files that are gonna be re-encoded for broadcasting by live encoders most of the time, but I know that the final distribution guys are always trying to get the best possible result 'cause unlike people in the team I work with, they're gonna have to deliver to the actual consumers, so they can't have the "luxury" to be like: "Oh, UHD, H.264, 500 Mbit/s 10bit, that will do" like the files I create but they're gonna have to find the best possible way to squeeze all that in a 25 Mbit/s H.265 stream encoded live.

Anyway, I'm digressing, the thing is that you're right, there's still a big market besides the consumer-niche.

[Jamaika]

Not really, given how crowded satellites like HotBird are and that many many many TVs all over the globe rely on them to broadcast their contents, it makes sense to keep the development of new codecs ongoing.
Compare the markets from 20 years ago. How many satellites have come and changed?
I've been to orange recently. You must have TV decoder, this will lower your bills. How does a TV set-top box work? Yes, the decoder is h264 but if you don't have optical fiber for high bitrates, you won't watch TV. What low data rate and why did you buy 16K TV? Is this data downsizing on satellites?
Last, how does this relate to the x265 codec. It's a codec used in simple converters.

[FranceBB]

Compare the markets from 20 years ago. How many satellites have come and changed?

Not many have been launched cause it's crazily expensive to do that.


I've been to orange recently. You must have TV decoder, this will lower your bills. How does a TV set-top box work? Yes, the decoder is h264 but if you don't have optical fiber for high bitrates, you won't watch TV.

I'm not familiar with orange and their decoders, but most set top box work with satellite dishes and rely on internet only for on demand stuff. In other words, any time you wanna watch a linear channel, you're gonna use the satellite dish. Now, when it comes to transmissions, almost every TV has MPEG-2 for SD, H.264 for FULL HD and H.265 for UHD. In our case it's around 4 Mbit/s for SD, 12 Mbit/s for FULL HD and 25 Mbit/s for UHD but as you probably are aware of, given that no new satellites have been launched and that literally the overwhelming majority of EU TVs are all on Hotbird 13B/13C/13E at 13.0°E whose bandwidth is the same as when the satellites have been originally launched, you can see why getting the lowest possible bitrate is important.
Besides, to those who say that streaming is the future, sure, granted, but not everyone has reliable broadband everywhere, especially those living in rural areas and that's where satellite comes in handy, but even then the biggest limiting factor is the cost: everyone wants UHD channels and there's just not enough bitrate in the satellites to grant it to everyone.





What low data rate and why did you buy 16K TV? Is this data downsizing on satellites?


Consumers will always want the latest and greatest technologies. If you stick your head in the sand and play dumb, you're never gonna gain new customers or indeed keep the ones that you already have.
Think about it: if no broadcaster updated its infrastructure, we would have been stuck with SD which was no longer competitive. People were buying HD TVs and wanted to watch HD contents. Sure, the bitrate was a problem then like it is now, but broadcasters had to cope with it and begin airing in HD. Then it came FULL HD, adding more struggle to the infrastructure. Right now, it's all about UHD, HDR 50p H.265 which is the standard. You know, imagine a world in which people bought UHD TVs and broadcasters kept airing in FULL HD BT709 SDR 25i or 30i H.264. People would just stop watching TV altogether or indeed complain about it and about the fact that they bought a new shiny TV and that their favourite sport is in FULL HD. And the circle goes on and on and on.



Last, how does this relate to the x265 codec. It's a codec used in simple converters.


I think you're by far underestimating how good x26x have become. This is not 2003 when there were dvix/xvid used by private consumers to re-encode their legally purchased MPEG-2 DVDs, this is 2022 where x264 and x265 are at the heart of contents encoded by companies all over the world, especially VOD.

[Barough]

x265 v3.5+39-58f46fa4a
Built on July 06, 2022, GCC 11.2.0

https://bitbucket.org/multicoreware/x265_git/commits/branch/master

DL:
https://www.mediafire.com/file/wc1t5q2ckc6390j/

Note :

The commit value is wrong due to something upstream at MulticoreWare.

Correct commit number for this release is
20255e6

[benwaggoner]

Last, how does this relate to the x265 codec. It's a codec used in simple converters.
x265 absolutely gets used in enterprise products and for enterprise-scale content. It's not as dominant as x264 is for H.264 encoding (where pretty much everyone but Beamr uses it under the hood), but still a very big deal.

[Jamaika]

Testing of new paid SAT functions LCEVC-X265. We don't have to go to the sat fair.

ffmpeg -y -i "Football.mp4" -g 60 -c:v lcevc_hevc -base_encoder x265 -r
29.97 -s 1920x1080 -b:v 0k -eil_params
"preset=veryslow;rc_pcrf=33;scenecut=0;min-keyint=60;frame-threads=4;residual_mode_priority_enabled=0;temporal_use_priority_map=0"
football_pCRF33_LCEVC_x265.mp4