Main Concept Hybrid GPU HEVC Encoder

[sophisticles]

I just ran across this:

https://www.mainconcept.com/products/for-developers/video/hevc-h265-codec.html?gclid=E...SAAEgJfy_D_BwE

They claim it's 20% more efficient than open source according to MSU but the real interesting thing is that they seem to have a hybrid GPU/CPU HEVC encoder, that runs on NVIDIA's RTX video cards, it seems to use software based bit rate and quality algorithms with GPU powered encoding, in other words not the NVENC encoder but a combination of CUDA and their software encoder, which they say is 2.5x as fast as their software encoder.

If someone with an RTX that run Windows could request a demo, which they offer, that would be great.

[sophisticles]

So I downloaded the sdk and sample applications for Linux and found some very interesting things, namely it appears it's at least partially based on open source software. Main Concept's HEVC encoder supports CRF; unfortunately checking through the included sample enc binary, it doesn't look like it supports the hybrid gpu mode in Linux.

If anyone on Windows is willing to try, it may prove very interesting.

[ibbanez]

So I see this is back from July, however I requested the SDK as well as the FFMPEG plugins, and wanted to check this out. I'm really curious about the Hybrid mode and its quality vs Speed vs x265. Is there anything, anyone wanted tested? I have a 2080ti also, by the way to test the Hybrid mode. Im not huge in this scene, but want to learn. Let me know what you guys would like to see tested.

[sophisticles]

Anything from http://ultravideo.cs.tut.fi/#testsequences that's 50fps.

[ibbanez]

Alright, so I downloaded some, and it seems to be running ok, but I'll be honest, I don't know what settings you guys are curious in. Now, Basically I downloaded their total code studio in order to make ini files, but they really don't have many options at all compared to x265. I've tried Variable bitrate and I ran CQ at Better once. But, if you have some specific settings you would like to see, I could try that, so that you can get a better look. This is all new to me, so I apologize for not being well versed in this, but I'm willing to try. So let me know what you guys would like to see, and where to upload the files, because I don't know how to compare them other than look, and they seem ok to me, but I don't know what I'm looking for. Thanks.

[ibbanez]

So update, if I use a regular file, the codec run great, and greatly reduces the file size, and things look awesome, and it is so fast. However, when trying to use the raw video from the site you posted, Im getting glitches. Right now, it messes up everytime. The file size is way to small, the output is all green, and I get constant Non-monotonous DTS in output stream 0:0; errors. I'm sure this is because I don't know what Im doing. I also had to add the -s 3840x2160 or else i got the invalid picture size error, but only when dealing with the raw video. now, i don't know enough about ffmpeg, or running all this manually, as I said before, I am a noob. Can you send me what you used for command line when you tested? And were you using the RAW 10-bit YUV files as well? If I use the other files, the 120fps that is ran at 23.98 fps and are MP4's, it runs without an issue, but that is already lower quality, and I wanted to use the RAW video for the best possible encodes. Im sorry I am new to this and don't know what I am doing, so any help would be great. But, for an untrained eye, the results look promising. The quality seemed adequate for viewing to me, but I also don't know how to tell if the results are perfect from an archival sense. Thanks for your help guys.

[ibbanez]

Got it mostly figured out I think. I have a couple of files to upload. Let me know where to upload them and link them if you are interested.

[sophisticles]

You can upload up to 500 mb per file directly to this site; I used ffmpeg to convert those files to y4m before I used them.

[ibbanez]

Ahh ok, I converted to mkv, and I also used their ffmpeg to convert them. I didn't like their full on encoder, so I used their ffmpeg plugin, but Im also new to all this. I'll upload my first two test files. One was with Constant Quantizer at Better, and the other was using CRF. Although I did find out CRF isn't available on 10 bit, only 8bit, so that sucks. Now the files sizes are ridiculously small, but I also don't know what I'm doing, but they look good to me. I'll add them here in one sec.

[ibbanez]

Here they are...

[ibbanez]

Now, Im still waiting to hear back from my email about how to control CRF, because inside all the settings file I created from Total Code Studio, I don't see the CRF Rate Option, so Im not sure, I just went with its default. Also, Q/P was set to 15 for these test.

[ibbanez]

So setting Rate_Factor = <int> does nothing apparently. Hopefully they write me back with some answers to my questions. I wish this wasn't a paid codec, but performance seems ok. I just need people with better skills than I, to tear these encodes apart and compare them, because I haven't the foggiest clue what I'm looking for. I don't see any blocking on initial looking.

[ibbanez]

Also, these Videos are wierd. They list 120fps, but if I select that, they are choppy as hell. If I select 23.98..., then they run smooth. Another detail I had issues with, the details also say they are 4:4:4, but if I select that, the color is all over the place, but if I select 4:2:0, then they are perfect. I think some of them are mislabeled, or Im just a dunce.

[ibbanez]

Figured out the Rate_factor issue I was having. Stupid error on my part. Gonna retest in a bit.

[Truthler]

Any news? Is it better than NVENC ?

[ibbanez]

I don't know how to quantitatively test these videos to be honest, that is why I was just trying to help out. There is also still things I'm learning and testing about the codec. I finally figured out the CRF issue, but haven't gotten back around to testing it yet, but will soon. I was just trying to add to the guys that started as they know a lot more about this stuff in general. I'd like to know what standard settings to use, so we can have as close to an even playing field when comparing. This codec did work quickly, but I need to test regular x265 with the same videos and see. I'm new, so I am doing what little I can to try and help. I don't know anything about the SMIM or PSNR stuff, or whatever they are called, only what my crappy eyes can see. But I'm interested in learning this stuff so I can be better informed.

[ibbanez]

Here are the CRF files from the encoder from 18-28 (the default CRF value). I hope to get some of the specialist input on what they think about these. Thanks.

[Truthler]

Here are the CRF files from the encoder from 18-28 (the default CRF value). I hope to get some of the specialist input on what they think about these. Thanks.

Do you use the full version?

What are the restrictions of the demo version?

[Truthler]

I don't know how to quantitatively test these videos to be honest, that is why I was just trying to help out. There is also still things I'm learning and testing about the codec. I finally figured out the CRF issue, but haven't gotten back around to testing it yet, but will soon. I was just trying to add to the guys that started as they know a lot more about this stuff in general. I'd like to know what standard settings to use, so we can have as close to an even playing field when comparing. This codec did work quickly, but I need to test regular x265 with the same videos and see. I'm new, so I am doing what little I can to try and help. I don't know anything about the SMIM or PSNR stuff, or whatever they are called, only what my crappy eyes can see. But I'm interested in learning this stuff so I can be better informed.

A simple visual video comparison software, it is not SSIM or PSNR based, just visual.
https://medium.com/the-svt-tech-blog/vivict-vivict-video-comparison-tool-7f6359f40283

Can you compare it with x265 at the same bitrate?

Thank you!

[ibbanez]

I believe the only restriction is the watermark that it adds to it. No I haven't purchased it, because I'm not sure I want to spend $100 on it or not. Yeah I can encode it in x265 and I'll check out this tool.. It may be a little bit, as Im getting ready for CyberPunk 2077 ;p stay tuned.

[Truthler]

I believe the only restriction is the watermark that it adds to it. No I haven't purchased it, because I'm not sure I want to spend $100 on it or not. Yeah I can encode it in x265 and I'll check out this tool.. It may be a little bit, as Im getting ready for CyberPunk 2077 ;p stay tuned.

The only SSIM and PSNR measurement tool is created by MSU. But it is not free . https://www.compression.ru/video/quality_measure/vqmt_download.html

[Truthler]

I believe the only restriction is the watermark that it adds to it. No I haven't purchased it, because I'm not sure I want to spend $100 on it or not. Yeah I can encode it in x265 and I'll check out this tool.. It may be a little bit, as Im getting ready for CyberPunk 2077 ;p stay tuned.

Any news?

Which settings have you used in the Mainconcept encoding and in the simple X265?

Do you have png pictures from the video comparison software?

[ibbanez]

so I can't apparently load h265 files in their test software, for the free one you listed. Only on Safari and I don't have a Mac, which seems dumb to me, but I digress. I tried Downloading the MSU free version, but that only does SD content. So I have no way of comparing these things. I've added the files up above, if you know how to, be my guest to download them and post the results. As soon as I get some spare time, I'll post the x265 version. Whats hard for me is, I'm new to this, so I don't have some standard set up settings that I use overall which I would think you would want as a baseline and to do across the board. The hybrid settings file looks like it has a lot of things to change, but realistically, there are only a handful. It doesn't have anywhere near the amount of options that x265 has, which I expected. Now, I will do what I normally do for x265, and if anyone wants any specific settings, I would be willing to do that, but for now, I am trying my best.

[ibbanez]

ok, noob question, but is there anyway to use Staxrip to load up the stock YUV raw video? That's what I normally use for my encoder, and I have my preset saved there. but its looking like I may have to switch to FFMPEG, but I didn't want to ;p If anyone knows a way to load it up in Staxrip, i would be grateful. Thanks.

[poisondeathray]

ok, noob question, but is there anyway to use Staxrip to load up the stock YUV raw video? That's what I normally use for my encoder, and I have my preset saved there. but its looking like I may have to switch to FFMPEG, but I didn't want to ;p If anyone knows a way to load it up in Staxrip, i would be grateful. Thanks.

raw YUV can be loaded into staxrip with an avisynth script ( rawsourceplus) or vapoursynth script (vsrawsource). If you're new, it might not make sense to you, a bit of a learning curve getting started. But ffmpeg might not make sense to you either - for ffmpeg a raw YUV source must have the parameters (-pix_fmt, dimensions -s, framerate -r) entered correctly before the -i input . If there is a choice to use Y4M instead of raw YUV, that's easier, because those parameters are conveyed by the Y4M headers

metrics such as PSNR/SSIM/VMAF etc...can be calculated with ffmpeg too. But they are less useful if a video has an added watermark, because that is obviously something different than the original and that watermarked video would be unfairly penalized . Depending on where the watermark is placed, one way around that might be to crop out the area from the original, and the test encodes , then compare, but it's still not ideal

[ibbanez]

Alright, I just simply ran the following for x265...

ffmpeg -f rawvideo -s 1920x1080 -pix_fmt yuv420p -i Jockey_1920x1080_120fps_420_8bit_YUV.yuv -c:v libx265 -preset slow -x265-params crf=18rofile=main10utput-depth=10 Jockey_1920x1080_120fps_420_8bit_YUV-x265-CRF18-test-results.mkv

And i'm uploading here. For comparison's, I will leave that to you guys, because I have no clue.

[ibbanez]

ok, noob question, but is there anyway to use Staxrip to load up the stock YUV raw video? That's what I normally use for my encoder, and I have my preset saved there. but its looking like I may have to switch to FFMPEG, but I didn't want to ;p If anyone knows a way to load it up in Staxrip, i would be grateful. Thanks.

raw YUV can be loaded into staxrip with an avisynth script ( rawsourceplus) or vapoursynth script (vsrawsource). If you're new, it might not make sense to you, a bit of a learning curve getting started. But ffmpeg might not make sense to you either - for ffmpeg a raw YUV source must have the parameters (-pix_fmt, dimensions -s, framerate -r) entered correctly before the -i input . If there is a choice to use Y4M instead of raw YUV, that's easier, because those parameters are conveyed by the Y4M headers

metrics such as PSNR/SSIM/VMAF etc...can be calculated with ffmpeg too. But they are less useful if a video has an added watermark, because that is obviously something different than the original and that watermarked video would be unfairly penalized . Depending on where the watermark is placed, one way around that might be to crop out the area from the original, and the test encodes , then compare, but it's still not ideal

Just saw this after I posted my upload ;p thanks though. Good to know about the AVS and VPS and I'll try it.

[Truthler]

ok, noob question, but is there anyway to use Staxrip to load up the stock YUV raw video? That's what I normally use for my encoder, and I have my preset saved there. but its looking like I may have to switch to FFMPEG, but I didn't want to ;p If anyone knows a way to load it up in Staxrip, i would be grateful. Thanks.

raw YUV can be loaded into staxrip with an avisynth script ( rawsourceplus) or vapoursynth script (vsrawsource). If you're new, it might not make sense to you, a bit of a learning curve getting started. But ffmpeg might not make sense to you either - for ffmpeg a raw YUV source must have the parameters (-pix_fmt, dimensions -s, framerate -r) entered correctly before the -i input . If there is a choice to use Y4M instead of raw YUV, that's easier, because those parameters are conveyed by the Y4M headers

metrics such as PSNR/SSIM/VMAF etc...can be calculated with ffmpeg too. But they are less useful if a video has an added watermark, because that is obviously something different than the original and that watermarked video would be unfairly penalized . Depending on where the watermark is placed, one way around that might be to crop out the area from the original, and the test encodes , then compare, but it's still not ideal

http://ultravideo.cs.tut.fi/#testsequences

Here is the webpage , where you can download uncompressed 4K yuv files in various formats (8 bit or 10bit)
I can't understand why did they use YUV container instead of AVI with losssless UT VIDEO encoder. It would be mathematically the same quality and 1/2 or 1/3 size of the original YUV files.

Can you recommend a software which can transform these into AVI or MKV files?

[poisondeathray]

http://ultravideo.cs.tut.fi/#testsequences

Here is the webpage , where you can download uncompressed 4K yuv files in various formats (8 bit or 10bit)
I can't understand why did they use YUV container instead of AVI with losssless UT VIDEO encoder. It would be mathematically the same quality and 1/2 or 1/3 size of the original YUV files.

Can you recommend a software which can transform these into AVI or MKV files?

YUV does not have a container . Y4M , if that is a given option - that is almost always better. Think of Y4M as the container for uncompressed YUV. It's easier to use because it has information on resolution, framerate, colorspace and sampling. If you pick raw YUV, you have to manually enter those values in the receiving software

Academic testing uses YUV or Y4M because almost all reasearch software use YUV/Y4M exclusively for encoding/decoding/analysis

You can use ffmpeg (or use one of the ffmpeg GUI's) to transform just about anything

e.g encode to UT video (note that the ffmpeg version only supports 8 bit UT video)

ffmpeg -i 8bit_source.y4m -c:v utvideo -an 8bit_utvideo_output.avi

If you wanted 10bit , you could use FFV1

ffmpeg -i input.y4m -c:v ffv1 -an output.mkv

[Truthler]

ok, noob question, but is there anyway to use Staxrip to load up the stock YUV raw video? That's what I normally use for my encoder, and I have my preset saved there. but its looking like I may have to switch to FFMPEG, but I didn't want to ;p If anyone knows a way to load it up in Staxrip, i would be grateful. Thanks.

raw YUV can be loaded into staxrip with an avisynth script ( rawsourceplus) or vapoursynth script (vsrawsource). If you're new, it might not make sense to you, a bit of a learning curve getting started. But ffmpeg might not make sense to you either - for ffmpeg a raw YUV source must have the parameters (-pix_fmt, dimensions -s, framerate -r) entered correctly before the -i input . If there is a choice to use Y4M instead of raw YUV, that's easier, because those parameters are conveyed by the Y4M headers

metrics such as PSNR/SSIM/VMAF etc...can be calculated with ffmpeg too. But they are less useful if a video has an added watermark, because that is obviously something different than the original and that watermarked video would be unfairly penalized . Depending on where the watermark is placed, one way around that might be to crop out the area from the original, and the test encodes , then compare, but it's still not ideal

Just saw this after I posted my upload ;p thanks though. Good to know about the AVS and VPS and I'll try it.

Hello Ibanez!

I will shot some lossless quality videos for the test, and we can compare this hybrid encoder with the NVENC engine. OK?