What is current status for hardware H.265 encoding.

[Mister XY]

what do you mean with low power mode?
Detected avaliable features for hw API v1.35, HEVC, Constant QP (CQP)
RC mode o
10bit depth o
Fixed Func o
Interlace o
VUI info o
Trellis x
Adaptive_I x
Adaptive_B x
WeightP o
WeightB o
FadeDetect o
B_Pyramid o
+ManyBframes o
PyramQPOffset o
MBBRC x
ExtBRC x
Adaptive_LTR x
LA Quality x
QP Min/Max x
IntraRefresh x
No Deblock o
No GPB o
Windowed BRC x
PerMBQP(CQP) o
DirectBiasAdj x
MVCostScaling x
SAO x
Max CTU Size x
TSkip x
HEVC SAO is not supported on current platform, disabled.

[Yups]

It's called fixed function in QSVEnc. Intel calls it low power.


--fixed-func use fixed func instead of GPU EU (default: off)

[Mister XY]

with --fixed-func qsvenc has a crash, but i see in log
HEVC SAO is not supported on current platform, disabled.

[Yups]

SAO isn't supported by QSVEnc and that's why you get this message. This is unrelated to the FF/low power mode.

[Mister XY]

How good is the ryzen 5600g in hardware encode?

[RanmaCanada]

How good is the ryzen 5600g in hardware encode?

AMD has always sucked (https://obsproject.com/forum/resources/ultimate-encoder-quality-analysis-2020-nvenc-vs-amf-vs-quicksync-vs-x264.998/). The 5600G runs Vega 7, and uses VCN 2.0. If you want to do hardware encoding, Intel Quicksync, especially the newest hardware, runs complete circles around it. Heck even Nvenc on Maxwell is better than AMD in some instances.

[Mister XY]

Ok then I know that the Ryzen is not good. Then I ask the question, what is the best APU / GPU for hardware encoding in x265 format?

[Yups]

CBR/VBR= Turing/Ampere
CQP= Iris Xe


Nivida has a big advantage with Lookahead in CBR/VBR mode but overall CQP offers best quality on Iris Xe.

[Mister XY]

That means that I currently have the best hardware for encoding because I don't use cbr or vbr.

[Yups]

It could be yes, although it's probably much slower than Iris Xe unless it's running in low power mode. I mean you can try and upload a sample, if you tell me the settings I can check out if it's the same. This (https://forum.doom9.org/showpost.php?p=1939967&postcount=437) or this (https://forum.doom9.org/showpost.php?p=1938769&postcount=422) you can use.

[Mister XY]

How do you check the vmaf?

[Yups]

I don't know. I'm using Video Quality Measurement Tool which isn't free. If the file size is identical there is no need for a quality metrics check.

[Mister XY]

OK, what for a sample do you need? A sample from surce ir from encode?

[Yups]

I need the encoded sample and settings you were using and driver version. If it's a known source for me I don't need that, so it depends.

[Yups]

Posting it here because some people might be interested. We made a comparison with the exact same settings between TGL-U Iris Xe and RKL-S UHD 750. It's a CQP best quality setting and software decoding.


File size/bitrate:

Iris Xe= 494.834 KB/8307.92 kbps
UHD750= 494.837 KB/8307.90 kbps


VMAF:

Iris Xe= 97.358
UHD 750= 97.358


Very minor file size/bitrate difference and same quality. From this I can say it's the same encoder which was expected but good to know it's the same. I was thinking the minor bitrate difference is decoding related. Another run with hardware decoding:


Iris Xe= 493.277 Kb
UHD 750= 493.286 Kb

Minor size difference as well. Iris Xe has two decoder unlike UHD750, this minor size difference could still related to decoding somehow. Doesn't matter in the end, same quality. Speed difference wasn't that big by the way.

[Yups]

From the release notes of Intels latest driver:

Support for H264 and HEVC DX12 video encode on Microsoft Windows® 11 for 11th Generation Intel® Core™ Processors.
https://downloadmirror.intel.com/648245/ReleaseNotes_100.9864.pdf

What is DX12 video encode and why Windows 11 only?

[nevcairiel]

Windows 11 adds a encoding API to DX12, which would be interesting as a cross-vendor way to access video encoding hardware, but the documentation on it seems to still be very much work in progress, so we'll see how usable that is once someone figures out how to use it.

[Balling]

From the release notes of Intels latest driver:


https://downloadmirror.intel.com/648245/ReleaseNotes_100.9864.pdf

What is DX12 video encode and why Windows 11 only?

dx11va is old. directx11over12 decoder will allow to do dx12 only and already works in Chrome under a flag.

[Yups]

Windows 11 adds a encoding API to DX12, which would be interesting as a cross-vendor way to access video encoding hardware, but the documentation on it seems to still be very much work in progress, so we'll see how usable that is once someone figures out how to use it.


Microsoft released it, there is a blog post: https://devblogs.microsoft.com/directx/announcing-new-directx-12-feature-video-encoding/

And on github: https://github.com/microsoft/DirectX-Specs/blob/master/d3d/D3D12VideoEncoding.md

[tonemapped]

dx11va is old. directx11over12 decoder will allow to do dx12 only and already works in Chrome under a flag.Chrome barely operates correctly with AV1 :D One of the recent updates changed my flag and decided my Atom-powered HTPC would love 4K AV1 :(

[pacuro]

It's called fixed function in QSVEnc. Intel calls it low power.


--fixed-func use fixed func instead of GPU EU (default: off)

I would like to test FF mode but notice crush each time I start encoding with --fixed-func.


--------------------------- Video encoding ---------------------------

QSVEnc 6.03

R:\StaxRip-v2.10.0-x64\Apps\Encoders\QSVEnc\QSVEncC64.exe --avsdll R:\StaxRip-v2.10.0-x64\Apps\FrameServer\AviSynth\AviSynth.dll --codec hevc --quality best --bframes 14 --b-pyramid --open-gop --sar 64:45 --sao none --fixed-func --cqp 20:22:24 --qp-offset 2:4:8 -i "R:\ninjago s03e30 dvd_temp\ninjago s03e30 dvd.avs" -o "R:\ninjago s03e30 dvd_temp\ninjago s03e30 dvd_out.hevc"

--------------------------------------------------------------------------------
R:\ninjago s03e30 dvd_temp\ninjago s03e30 dvd_out.hevc
--------------------------------------------------------------------------------
HEVC SAO is not supported on current platform, disabled.
cop3.DirectBiasAdjustment value changed off -> auto by driver
cop3.GlobalMotionBiasAdjustment value changed off -> auto by driver
QSVEncC (x64) 6.03 (r2496) by rigaya, Sep 25 2021 05:36:09 (VC 1929/Win)
OS Windows 10 x64 (19043) [UTF-8]
CPU Info 11th Gen Intel Core i7-11700K @ 3.60GHz [TB: 4.00GHz] (8C/16T) <Tigerlake>
GPU Info Intel UHD Graphics 750 (32EU) 350-1300MHz [125W] (30.0.101.1191)
Media SDK QuickSyncVideo (hardware encoder) FF, 1st GPU, API v2.06
Async Depth 3 frames
Buffer Memory d3d11, 24 work buffer
Input Info AviSynth+ 3.7.0 r3382(yv12)->nv12 [AVX2], 720x576, 25/1 fps
AVSync cfr
Output HEVC(yuv420) main @ Level 3
720x576p 64:45 25.000fps (25/1fps)
Target usage 1 - best
Encode Mode Constant QP (CQP)
CQP Value I:20 P:22 B:24
QP Limit min: 10, max: 51
Trellis Auto
Ref frames 5 frames
Bframes 14 frames, B-pyramid: on
Max GOP Length 250 frames
Ext. Features WeightP WeightB QPOffset
MFXENCODE: EncodeFrameAsync error: device operation failure..
Break in task MFXENCODE: device operation failure..
encoded 2 frames, 1.53 fps, 2608.80 kbps, 0.02 MB
encode time 0:00:01, CPULoad: 0.1
frame type IDR 1
frame type I 2, total size 0.05 MB
QSVEncC.exe finished with error!


I have noticed --fixed-func sets QP Limit min: 10 when EU encoding mode sets QP Limit min: 1.
Is it important? How to enable FF encoding properly?

[Yups]

Try out the newest driver, if it doesn't help you need a bios update if there is a new one available. There was a bug on RKL-S which affected FF encoding: https://github.com/Intel-Media-SDK/MediaSDK/issues/2784

QP Limit is not important, it can't be changed manually anyways.

[pacuro]

Try out the newest driver, if it doesn't help you need a bios update
Both updated right now. Still getting error. Thanks for hint anyway. Now I know there is global problem with hevc ff encoding on rkl. Hope folks are working on that in intel.

UPDATE: Asrock prepared alpha bios with some fix which did not help. I have replaced microcode to last version "50" for A0671 and FF encoding started working. 320 fps processing 1080p video.

[lt8nk]

Hello everyone,
I am reading this post since few months because I would like to upgrade my actual configuration and this is the only forum where people are talking about QuickSync and its quality in hardware encoding. Here most people are talking about UHD750. But do you know if UHD710 (Pentium G7400T) is also able to obtain the same result ?

By the way, I saw a post asking how to get the VMAF value. Here is a software that calculates it : https://github.com/fifonik/FFMetrics

Thank you again for all these informations

[Yups]

In the worst case GT1 graphics only supports the fully fixed function mode, otherwise it should have the same quality. It's the same architecture in the end.

[lt8nk]

GT1 is Haswell. I don't understand. But what would be the consequences of only supporting the full fixed function ?

Sorry, even if I read the thread, I may have silly questions.

[Yups]

For example the slower Atom from Intel supported Quicksync but only the fixed function encoding and not the shader supported Hybrid encoding from the faster Core lineup. I don't know if this is the case for UHD710 with only 16EUs, just saying this is the worst case. You can check my posting history, I posted several VMAF tests with FF and hybrid. Sure Hybrid has a better quality, although the difference isn't really big, however the speed difference is huge. Hybrid encoding might not even make sense therefore.

[lt8nk]

Thank you. I found your post (https://forum.doom9.org/showpost.php?p=1940526&postcount=451). I remember it now. I read it before. Hybrid is better but FF should be enough. Right now I am encoding in x265 medium with a Ryzen 1500x and it takes A LOT of time.

I used to hesitate with the 12100T but I'll try the G7400T or the G7400 as soon as the price of the 1700 motherboards goes down a bit. But I think it will not be for a few months. And I will come back here to share my tests with you.

[RanmaCanada]

Thank you. I found your post (https://forum.doom9.org/showpost.php?p=1940526&postcount=451). I remember it now. I read it before. Hybrid is better but FF should be enough. Right now I am encoding in x265 medium with a Ryzen 1500x and it takes A LOT of time.

I used to hesitate with the 12100T but I'll try the G7400T or the G7400 as soon as the price of the 1700 motherboards goes down a bit. But I think it will not be for a few months. And I will come back here to share my tests with you.

Why not just upgrade your processor to a zen3 like a 5600? You will get a significant increase in encode speed without having to spend much money on a new platform.

[perrinpages]

What is the quality of the Apple M1 based HW encoders? I haven't seen any real tests or comparisons done.

[RanmaCanada]

What is the quality of the Apple M1 based HW encoders? I haven't seen any real tests or comparisons done.

From reading on the macforums (https://forums.macrumors.com/threads/mac-mini-m1-h-265-encoding.2269815/), it's garbage tier as it's even worse than Pascal NVENC.

[perrinpages]

From reading on the macforums (https://forums.macrumors.com/threads/mac-mini-m1-h-265-encoding.2269815/), it's garbage tier as it's even worse than Pascal NVENC.

Meh. I don't trust them, no real evidence. They all seem to be using handbrake...I would rather see results using compressor.

[RanmaCanada]

Meh. I don't trust them, no real evidence. They all seem to be using handbrake...I would rather see results using compressor.

I doubt changing the software would make that big of a difference when we're talking about hardware encoding as the API should be pretty much the same across all platforms. But I don't use Mac's so I could quite possibly be very wrong. Take it or leave it.

[Ritsuka]

I remember that it was mostly tuned for psnr, and right, unless Apple hided a secret "enable better quality" switch in Compressor, HandBrake will have the same exact output.

[perrinpages]

I doubt changing the software would make that big of a difference when we're talking about hardware encoding as the API should be pretty much the same across all platforms. But I don't use Mac's so I could quite possibly be very wrong. Take it or leave it.

Handbrake's settings are fussy and the defaults are not sane at all. Yes, Handbrake and Compressor both use the videotoolbox framework, but we haven't seen any jsons or logs so we will never know if their conclusions on the quality are purely encoder based or if it's handbrake's filtering.

[Ritsuka]

VideoToolbox has got exactly zero useful settings, I highly doubt I screwed something up when I implemented VideoToolbox support in HandBrake.
But if you know how to improve it, feel free to point it out.

[RanmaCanada]

Handbrake's settings are fussy and the defaults are not sane at all. Yes, Handbrake and Compressor both use the videotoolbox framework, but we haven't seen any jsons or logs so we will never know if their conclusions on the quality are purely encoder based or if it's handbrake's filtering.

Well at this point, since you're being so difficult, maybe you should buy an M1 Mac and report back as it appears nothing will satisfy you unless you do it yourself. Logs are very rarely even posted here, and most people requesting said comparisons don't go to the level you are in regards to demanding proof.

Sorry but no one cares about Macs here it appears, so you'll need to do the legwork that others did in this thread.

[perrinpages]

VideoToolbox has got exactly zero useful settings, I highly doubt I screwed something up when I implemented VideoToolbox support in HandBrake.
But if you know how to improve it, feel free to point it out.

Your implementation is fully optimized for HDR?

[perrinpages]

Well at this point, since you're being so difficult, maybe you should buy an M1 Mac and report back as it appears nothing will satisfy you unless you do it yourself. Logs are very rarely even posted here, and most people requesting said comparisons don't go to the level you are in regards to demanding proof.

Sorry but no one cares about Macs here it appears, so you'll need to do the legwork that others did in this thread.

Why are you so pressed? I asked a question and you replied with a useless macrumors forum that I've already read. In my first post I asked if someone had done real comparisons. Anyone who claims 2.5 Mb/s using x265 is an acceptable bitrate for HD content cannot be trusted to make quality assessments.

[BuccoBruce]

On Intel both HD630 and Iris Xe the constant bitrate modes seem to work a lot better with a low bitrate budget. For now only a PSN-Y test using this (https://drive.google.com/file/d/1YX1V0SeSkYaq6Ui41vv1wcOatbLnuLSL/view?usp=sharing) sample, all of them HEVC main 8 bit 1920x1080 at ~2.5 Mbit 24 fps/240 gop. varna, you might use this sample.
/snip/
https://s10.directupload.net/images/201212/cs73djrc.png

The difference is really big, Gen9 looks much worse in the video.

Another x265/x265 CRF vs Iris Xe CQP comparison using this (https://drive.google.com/file/d/1YX1V0SeSkYaq6Ui41vv1wcOatbLnuLSL/view?usp=sharing) sample.
/snip/


Would I be mistaken in assuming this would be the easiest sample to provide AMD VCE results for, for comparison's sake? Once I figure out a way to streamline running multiple encodes and then testing those metrics, and have enough time to play with settings.

I flipped through a few pages and didn't see any AMD results - or any single test clip with as broad a range of encodes/settings (thank you Yups). I have older Polaris Radeon/Radeon Pro WX cards with VCE 3.4 (technically Lexa, Ellesmere, and "Polaris 20" - they should differ only in encode speed, I'll double check VCEEnc's output but will most likely use the Ellesmere card) and a Radeon RX 5700 with VCN 2.0. I was always under the impression that while AMD's HW H.264 encodes were kind of rubbish, H.265 was "OK".

[RanmaCanada]

Why are you so pressed? I asked a question and you replied with a useless macrumors forum that I've already read. In my first post I asked if someone had done real comparisons. Anyone who claims 2.5 Mb/s using x265 is an acceptable bitrate for HD content cannot be trusted to make quality assessments.

Because over the almost decade that this thread has been alive, no one has mentioned Mac's at all. Common sense would dictate that obviously no one here cares for them (they aren't mentioned in any threads that I remember), or sees them as viable. You also did not mention that you had read the forums that I posted. I guess that means you've also watched this video? (https://www.youtube.com/watch?v=RzhvdRisNZM) and his subsequent video on it.

Sorry for trying to give you some information about a subject no one here cares about.

[Ritsuka]

That video has got nothing to do with the hardware encoder, and like 99% of YouTube videos is probably a bunch of half-truth and comparing apples to oranges.

Anyway, like I said there should not be any difference between HandBrake a Compressor. There is some data on https://forum.handbrake.fr/viewtopic.php?p=193484#p193484 , I don't know any more recent and decent data on this topic.

[Mister XY]

Here are my settings for HB. I use ist for UHD sources. But it's look not bad with FHD sources. Not Perfekt, but not bad for this time.

[Yups]

Would I be mistaken in assuming this would be the easiest sample to provide AMD VCE results for, for comparison's sake? Once I figure out a way to streamline running multiple encodes and then testing those metrics, and have enough time to play with settings.

I flipped through a few pages and didn't see any AMD results - or any single test clip with as broad a range of encodes/settings (thank you Yups). I have older Polaris Radeon/Radeon Pro WX cards with VCE 3.4 (technically Lexa, Ellesmere, and "Polaris 20" - they should differ only in encode speed, I'll double check VCEEnc's output but will most likely use the Ellesmere card) and a Radeon RX 5700 with VCN 2.0. I was always under the impression that while AMD's HW H.264 encodes were kind of rubbish, H.265 was "OK".


I don't have AMD. Yes this sample is easy to compare.

Here are my settings for HB. I use ist for UHD sources. But it's look not bad with FHD sources. Not Perfekt, but not bad for this time.


Unfortunately Handbrake doesn't support open gop yet which improves the bitrate efficiency a bit. It is planned though.

[BuccoBruce]

I don't have AMD. Yes this sample is easy to compare.

Thank you! I was offering to add to your already extensive results. Looks like ffmpeg might be able to output simple PSNR/SSIM/VMAF scores, all of the pretty GUI tools I can find seem to be expensive, professional, or both, and that Intel one looks like it's been discontinued. The MSU one would be ideal if it wasn't limited to 720p...

I'll try and get results up in a few days or so, with added 10-bit output from the Navi card since it supports it.

[Yups]

If you upload the encoded sample I can calculate it with MSU, otherwise it's not comparable to my scores.

[BuccoBruce]

Here's a preliminary look at how AMD fares with the Intel test clip, and it doesn't look too good. CQP 22:24 is VCEEncC's default, and you need RDNA2 for B-frames, but I'm not sure they'd help. Pre-analysis isn't supported for HEVC. There's a "Pre-Encode" rate control feature, but I'm not sure it'll help either. So here's "Default" for now, with just changing the preset from balanced to either slow or fast.

The "higher end" GPU produced larger files across the board, with the same exact settings. "Balanced" and "Slow" seem to produce identical output on both cards!


Quality/GPU VMAF PSNR SSIM Speed Bitrate Size
Bal/WX2100 95.3114 42.5086 0.9782 147.43 fps 7096.70 kbps 412.70 MB
Bal/WX5100 96.1465 42.9197 0.9785 121.80 fps 7963.64 kbps 463.12 MB
Fast/WX2100 95.2958 42.5546 0.9783 152.93 fps 7158.12 kbps 416.28 MB
Fast/WX5100 96.2284 43.1607 0.9798 122.83 fps 7890.48 kbps 458.86 MB
Slow/WX2100 95.3114 42.5086 0.9782 148.30 fps 7096.70 kbps 412.70 MB
Slow/WX5100 96.1465 42.9197 0.9785 121.26 fps 7963.64 kbps 463.12 MB


https://i.postimg.cc/SNKSL2X3/allVMAF.png (https://postimages.org/)

https://i.postimg.cc/gkBYgz8J/PSNR.png (https://postimages.org/)

https://i.postimg.cc/bwDqM79R/VMAF.png (https://postimages.org/)

"Lexa PRO GL" in the Radeon Pro WX 2100 is a Polaris 12 (https://www.techpowerup.com/gpu-specs/amd-lexa.g806) or "Lexa" GPU, and isn't very different from the RX 550. "Polaris 10 PRO GL (https://www.techpowerup.com/gpu-specs/amd-ellesmere.g795)" in the WX 5100 is Polaris 10, a single slot, toned down Radeon RX 480. They're both GCN 4.0 GPUs with VCE 3.4. I didn't expect them to perform differently, unless someone at AMD thinks ~15% is within the margin of error for a non-deterministic encode. I wouldn't be surprised if Polaris 20 XL (https://www.techpowerup.com/gpu-specs/radeon-rx-570.c2939) gave entirely different results too, even though it too reports as codename "Ellesmere" and also has VCE 3.4. Might try it later.

-------------------------------------------------------------------------------------
VCEEnc (x64) 7.00 (r1066) by rigaya, Apr 30 2022 18:34:01 (VC 1931/Win)
GPU: Radeon Pro WX 5100, AMF Runtime 1.4.23 / SDK 1.4.24
Input Info: AviSynth+ 3.7.2 r3661(yv12)->nv12 [AVX], 1920x1080, 24/1 fps
Vpp Filters copyHtoD
Output: H.265/HEVC main @ Level 4 (main tier)
1920x1080p 0:0 24.000fps (24/1fps)
Quality: slow
CQP: I:22, P:24
VBV Bufsize: 12000 kbps
Bframes: 0 frames
Pre Analysis: off
Motion Est: Q-pel
Slices: 1
GOP Len: 240 frames
VUI: matrix:bt709,colorprim:bt709,transfer:bt709
Others: deblock


The only change between the three runs was "Quality" which was either slow, balanced, or fast. Everything else was left at its default settings (for now). I might try another run with PE enabled, 2.5 Mbps VBR. And yes, I quadruple checked that the "Slow" and "Balanced" runs weren't mislabeled, I had log files from running the encodes. FFMetrics keeps crashing, although I did ask it to save CSV frame data if anyone cares for it. :logfile: I just can't play with the plots directly within it anymore without re-running an hour's worth of tests and I doubt it's worth opening Excel for.

Doesn't seem like RDNA2 fares much better either. (https://codecalamity.com/hardware-encoding-4k-hdr10-videos/)

[BuccoBruce]

If you upload the encoded sample I can calculate it with MSU, otherwise it's not comparable to my scores.

Whoops, I just saw that while I was writing that post. Here you go! The clips are still uploading https://mega.nz/folder/SzwDlR5D#SbfBKtLuD3rOQPjTUZ0P8g

[Yups]

The quality is quite poor considering it's converted into 8 Mbit. Bframes should make a noticeably difference, although the difference appears so big they need much more improvements.

[BuccoBruce]

The quality is quite poor considering it's converted into 8 Mbit. Bframes should make a noticeably difference, although the difference appears so big they need much more improvements.

Yeah, I'm fairly disappointed in how it turned out. I thought AMD VCE/VCN being rubbish was just a meme but I guess it's true. Almost tempted to dig a Sandy/Ivy Bridge CPU with QSV out and compare it to 8 Mbps AVC on that thing.