What is current status for hardware H.265 encoding.

[Yups]

I'm close to buying a cheap Intel mini PC for realtime 1080p H265 encoding.
I'm trying to decide between Gemini Lake N4020 and newer Jasper Lake N5105. Both have Intel UHD Graphics.

Considering both can do realtime 1080p H265 HW encodes, is there any reason to think that there would be a quality improvement using the newer processor?


Gemini Lake N4020 features a Gen9 GPU and Jasper Lake Gen11. Gen11 is better of course, it comes with an upgraded HEVC encoder.

The problem might be Gen9/Gen11 doesn't support bframes in fully fixed function mode. Icelake-U didn't have bframes support in FF which is Gen11 based. Unless Jasper Lake differs this isn't optimal because without bframes the quality will drop quite a bit, same for Gemini Lake. And maybe the hybrid mode won't work on Atom based SKUs anyways. If you can wait the upcoming ADL-N is better.


Can you please tell us the encoding speed?

Can you increase/decrease quality factor to see how the samples scale?


On this sample about 160 fps for AV1/HEVC QSV. This is a RAW sample, hardware decoding doesn't work there. Balanced preset isn't faster therefore, the encoding engine is not the limiting factor. I haven't checked x265 speed and I only have a quadcore CPU, this isn't too meaningful.

[tormento]

On this sample about 160 fps for AV1/HEVC QSV.
:eek:

Will the different Arc cards feature the same encoding speed (such as nvidia) or it will be dependent to the number of cores?

[Yups]

:eek:

Will the different Arc cards feature the same encoding speed (such as nvidia) or it will be dependent to the number of cores?



They have the same encoder/decoder. It can differ depending on the clock speed and memory speed differences if it isn't a fixed function encoder+decode. I would assume the speed is similar if it's a fully fixed function encode. There is a good test from techgage (https://techgage.com/article/intel-arc-a750-a770-workstation-review/). Similar speed in Handbrake (https://techgage.com/viewimg/?img=https://techgage.com/wp-content/uploads/2022/10/Intel-Arc-A770-and-A750-Performance-HandBrake-Transcode.jpg&desc=Intel%20Arc%20A770%20and%20A750%20Performance%20(HandBrake%20Transcode)) which I think is representative for a fully fixed function transcode. There are bigger differences in Vegas Pro and Adobe, hard to say why. ProRes encode in Adobe is certainly without hardware decoding, so maybe the faster memory from the bigger Arc cards helps. The other Adobe encode is with heavy effects, I guess it uses the GPU for it, in such a case the difference will be big.

[outhud]

Gemini Lake N4020 features a Gen9 GPU and Jasper Lake Gen11. Gen11 is better of course, it comes with an upgraded HEVC encoder.
Thanks for this info! I see the encoder has been upgraded to support higher resolutions, but I take from your post that it also has been upgraded to give better quality at 1080p?

If the 1080p quality is the same, then the N4020 will be my best choice because it beats N5105 on price.

1080p H265 is my only use case for the CPU, I'll be running it as a headless Linux server running only tvheadend to transcode digital TV to H265 in realtime.

[Yups]

Thanks for this info! I see the encoder has been upgraded to support higher resolutions, but I take from your post that it also has been upgraded to give better quality at 1080p?

If the 1080p quality is the same, then the N4020 will be my best choice because it beats N5105 on price.

1080p H265 is my only use case for the CPU, I'll be running it as a headless Linux server running only tvheadend to transcode digital TV to H265 in realtime.


There is big quality difference, I mean it's Gen11 vs Gen9.


Other improvements include a new HEVC Quick Sync Video engine that provides up to a 30% bitrate reduction over Gen9 (at the same or better visual quality)

For the media block, Intel says that the Gen11 design includes a ground up HEVC encoder design, with high quality encode and decode support.

https://www.tomshardware.com/reviews/intel-sunny-cove-gen11-xe-gpu-foveros,5932-3.html
https://www.anandtech.com/show/13699/intel-architecture-day-2018-core-future-hybrid-x86/3


Atom based Gen11 SKUs only support the FF mode, I don't think there is bframes support. Not ideal but of course better than Gen9 iGPU.

[RanmaCanada]

I'm close to buying a cheap Intel mini PC for realtime 1080p H265 encoding.
I'm trying to decide between Gemini Lake N4020 and newer Jasper Lake N5105. Both have Intel UHD Graphics.

Considering both can do realtime 1080p H265 HW encodes, is there any reason to think that there would be a quality improvement using the newer processor?

If you're in the states, why not look on ebay for an 11th gen laptop with a smashed screen (https://www.ebay.com/itm/394288111429?_trkparms=amclksrc%3DITM%26aid%3D1110006%26algo%3DHOMESPLICE.SIM%26ao%3D1%26asc%3D20201210111314%26meid%3D1aa20a900add40d1a93aa564e745e6d3%26pid%3D101195%26rk%3D1%26rkt%3D12%26sd%3D354323643789%26itm%3D394288111429%26pmt%3D1%26noa%3D0%26pg%3D2047675%26algv%3DSimplAMLv11WebTrimmedV3MskuAspectsV202110NoVariantSeedKnnRecallV1%26brand%3DLenovo&_trksid=p2047675.c101195.m1851&amdata=cksum%3A3942881114291aa20a900add40d1a93aa564e745e6d3%7Cenc%3AAQAHAAABIMFr2e4EmAnM%252ByHZkULYKDIJ4L66fOjNL0iupgt%252BzO1%252F3AE1t3mNirUYB96NktMCicMagiS6mbeTl0xquGODv9l8iKOqtJXlKg2qgQBwOmNjHsQkhJ1ju%252FQXecFbzwuG6WTigWFgGu25g4p5f5YcE0FWEhtU6tWalHRoLnVZuZXSqqrl%252BENED49wlbSc0AWxDfeNhhoDZGgYg4RAzRCQLbSZV%252BxBuNORsVLMUpOWy3eneCCFtkRnOzEnhObbacQ7fmTkWO2p3CLjd8t%252FZGKXFTOp9XGs1dWuIgY17TBfDE9nrUZFufYAKUgcHnEtZJ8kowM7tg%252BDHtyRuOq7WYdwoFofvXVEMvb0dyBBwBiIZKQC6HIZBJvKX076HtWSHz9h6w%253D%253D%7Campid%3APL_CLK%7Cclp%3A2047675) or refurbished (https://www.ebay.com/itm/275225296169?hash=item4014b4bd29:g:FXUAAOSwuvdjIiwU&amdata=enc%3AAQAHAAAAsB5ABgwTztmoBFS3MsPcV0tP%2FI%2B4vCJohV5fp2vqvDLCHTh1tb2ATcBRjZY1ij2ZJqTmdzkQyDHNqkV%2B2pkj3qJynAdNYNYinzBTnMTF0mb8NDkcNAtSWRQvadDA8r5w06Y1qodr%2F%2Fnpbq04fg4hmhXyS0zD1Kl%2BxIcREpXStvP9CXddklD%2FthAc5%2B%2FqOMvzTmV9ZkKrsy2Y4fNsyxb4omLl3aUd14UDC0VWJ0mcKJzS%7Ctkp%3ABk9SR8qTvO37YA)?

I picked up an i3-1115G4 laptop off eBay for $120 as it had a smashed screen, and I'm using it as my Emby/Plex server. External out works, and you can just install remote software like Anydesk or whatever you like, and Bob's your uncle.

[outhud]

There is big quality difference, I mean it's Gen11 vs Gen9.

https://www.tomshardware.com/reviews/intel-sunny-cove-gen11-xe-gpu-foveros,5932-3.html
https://www.anandtech.com/show/13699/intel-architecture-day-2018-core-future-hybrid-x86/3

Atom based Gen11 SKUs only support the FF mode, I don't think there is bframes support. Not ideal but of course better than Gen9 iGPU.

Great! Thanks a lot. I think the Gen11 wins out so after reading those links.

[tormento]

40xx series capabilities (https://github.com/rigaya/NVEnc/blob/master/GPUFeatures/rtx4090.txt)

[GrandPa]

Did anybody recently test the new Intel ARC A380 on Intel's 8th generation CPU platform?
Several reviews praised the video encoding capabilities of the A380, but on the other hand explicitely mentioned severe problems to get the card up and running on "old" platforms without rBAR feature enabled. Intel doesn't mention platforms below 10th generation CPUs in the compatibility list.

@YUPS: Which platform did you use?

Since nearly 3 years, I was using HEVC HW encoding on a GTX-1660 card (Turing chip) with remarkably high encoding speed -compared to SW encoding- and good quality and compression rate, mainly on 720p HDTV material. I'm using a i7-8700K CPU on a GA Z370-HD3P motherboard.
Now, I would like to give the ARC A380 with its HEVC and AV1 capabilities a try, but I'm wondering whether the card would run on my system.

Any remarks are welcome.

[RanmaCanada]

Did anybody recently test the new Intel ARC A380 on Intel's 8th generation CPU platform?
Several reviews praised the video encoding capabilities of the A380, but on the other hand explicitely mentioned severe problems to get the card up and running on "old" platforms without rBAR feature enabled. Intel doesn't mention platforms below 10th generation CPUs in the compatibility list.

@YUPS: Which platform did you use?

Since nearly 3 years, I was using HEVC HW encoding on a GTX-1660 card (Turing chip) with remarkably high encoding speed -compared to SW encoding- and good quality and compression rate, mainly on 720p HDTV material. I'm using a i7-8700K CPU on a GA Z370-HD3P motherboard.
Now, I would like to give the ARC A380 with its HEVC and AV1 capabilities a try, but I'm wondering whether the card would run on my system.

Any remarks are welcome.

Intel specifically states you need an rBAR system for the card to work. The reason they don't mention them, is because they "aren't compatible", and you're on your own if you attempt to use them.

https://www.intel.com/content/www/us/en/support/articles/000091128/graphics.html

It is quite possible the card will work on your older system, but Intel does not guarantee any type of performance.

https://www.techpowerup.com/review/intel-arc-a770-pcie-3-resizable-bar/ Sorry no encoding tests.

All you can do is buy it and try it, and report back.

[GrandPa]

Intel specifically states you need an rBAR system for the card to work. The reason they don't mention them, is because they "aren't compatible", and you're on your own if you attempt to use them.

https://www.intel.com/content/www/us/en/support/articles/000091128/graphics.html

It is quite possible the card will work on your older system, but Intel does not guarantee any type of performance.

https://www.techpowerup.com/review/intel-arc-a770-pcie-3-resizable-bar/ Sorry no encoding tests.

All you can do is buy it and try it, and report back.

Many thanks for the link with GPU-Z. It helped me to find out all the rBAR requirements and to discover that the BIOS of my MoBo actually does support rBAR (it was hidden behind the "Above 4G Decode" setting, which I hadn't enabled yet).
Thus, a new A380 card is ordered, let's wait and see. I will report about it later.

[Yups]

Did anybody recently test the new Intel ARC A380 on Intel's 8th generation CPU platform?
Several reviews praised the video encoding capabilities of the A380, but on the other hand explicitely mentioned severe problems to get the card up and running on "old" platforms without rBAR feature enabled. Intel doesn't mention platforms below 10th generation CPUs in the compatibility list.

@YUPS: Which platform did you use?

Since nearly 3 years, I was using HEVC HW encoding on a GTX-1660 card (Turing chip) with remarkably high encoding speed -compared to SW encoding- and good quality and compression rate, mainly on 720p HDTV material. I'm using a i7-8700K CPU on a GA Z370-HD3P motherboard.
Now, I would like to give the ARC A380 with its HEVC and AV1 capabilities a try, but I'm wondering whether the card would run on my system.

Any remarks are welcome.



I have an older 7th Gen CPU i7-7700k+Z170. I don't use it for gaming, although I think missing rBAR might effect encoding performance in some cases with bigger PCIe copy work, for example when it's not using hardware decoding.

[ukmark]

@Yups The MOBO on my Ice Lake laptop gave up the ghost, but I managed to get a great deal on an Acer i5-1135 G7 laptop.
It seems to me that FF encoding with this gives very good quality. As you say, Hybrid does give better quality, but how noticeable is the difference? I did a quick test here and I struggled to actually see any difference between Hybrid and FF for the same CQP values. The VMAF, PSNR and SSIM scores for Tiger Lake HEVC seem very close between FF and Hybrid from your earlier tests. Where do you see the differences in quality between FF and Hybrid, (is it facial details etc)?? Or, is it that the differences are so marginal, that FF is the way to go with Tiger Lake??

Here are my encoding options from StaxRip v2.13:-
--avhw --codec hevc --quality best --profile main10 --bframes 14 --b-pyramid --adapt-ltr --open-gop --async-depth 4 --no-repeat-pps --no-tskip --sao none --fixed-func --cqp 30:32:33 --qp-offset 2:4:8

TIA and BTW thanks for all your testing and advice!!

[Yups]

Hybrid has a little more details in direct screenshot comparisons and the file size is a little smaller but it's not a big difference. Because of the big speed difference FF is the preferred choice in most cases. You can try without the custom offset by the way, maybe it's slightly better. CQP with 14 bframes is really good for sure. I tested with FFMPEG the last days, it's really good. But there is no SAO option to turn it off.

[ukmark]

Hybrid has a little more details in direct screenshot comparisons and the file size is a little smaller but it's not a big difference. Because of the big speed difference FF is the preferred choice in most cases. You can try without the custom offset by the way, maybe it's slightly better. CQP with 14 bframes is really good for sure. I tested with FFMPEG the last days, it's really good. But there is no SAO option to turn it off.

Thanks for the reply. I really like the speed of FF with TGL. I'll try having a go with no qp offsets as you suggest and may also try using my original CQP settings but lowering the main CQP I/P/B settings to 29/31/32 from 30/32/33 and see if the size difference is significant. Thanks again. I just want a "set it and forget" preset that just works. If it means a slightly bigger file size, but same quality as Hybrid, then that's no problem.

[ukmark]

I did some more testing of fixed-function vs hybrid and for the life of me I can't see any differences. I used the same CQP settings for both encodes (the only difference being fixed-function was turned on in one encode and turned off in the other). I used the StaxRip tool "Video Comparison" which allows you to compare the exact same frames from multiple files, in this case just the two files. Although you can't see a side-by-side view, the tool has a tab at the top of the gui (one tab for each file). If you click on each tab back and forth, you can usually see any differences, as the screen just overlays the current frame from that video.

In my testing, it was like a mirror image. I tested a video which had a lot of dark and bright scenes, fast and slow movement and I just could not see any differences. I just dragged the slider in the gui to random parts of the video and must have looked at over 200 frames. Also when playing back both files, there was just no observable differences. So, to my eyes, Tiger Lake has made hybrid redundant, I'm just sticking with fixed-function encoding.

I know that AV1 will eventually take over from HEVC, but right now I am very happy with 10-bit HEVC fixed-function encodes on Tiger Lake.

[RanmaCanada]

I did some more testing of fixed-function vs hybrid and for the life of me I can't see any differences. I used the same CQP settings for both encodes (the only difference being fixed-function was turned on in one encode and turned off in the other). I used the StaxRip tool "Video Comparison" which allows you to compare the exact same frames from multiple files, in this case just the two files. Although you can't see a side-by-side view, the tool has a tab at the top of the gui (one tab for each file). If you click on each tab back and forth, you can usually see any differences, as the screen just overlays the current frame from that video.

In my testing, it was like a mirror image. I tested a video which had a lot of dark and bright scenes, fast and slow movement and I just could not see any differences. I just dragged the slider in the gui to random parts of the video and must have looked at over 200 frames. Also when playing back both files, there was just no observable differences. So, to my eyes, Tiger Lake has made hybrid redundant, I'm just sticking with fixed-function encoding.

I know that AV1 will eventually take over from HEVC, but right now I am very happy with 10-bit HEVC fixed-function encodes on Tiger Lake.

I am going to have to agree with you. In my limited testing, I encoded the same 4k movie, and I could not tell the difference between x265 and my Tiger Lake encode, other than the speeds. Tiger Lake managed 15+fps, while my x265 encode was going at around 1-1.5. I think it's time I get a larger ssd for my laptop so I can start encoding more stuff directly instead of dumping it on the network.

[ukmark]

I am going to have to agree with you. In my limited testing, I encoded the same 4k movie, and I could not tell the difference between x265 and my Tiger Lake encode, other than the speeds. Tiger Lake managed 15+fps, while my x265 encode was going at around 1-1.5. I think it's time I get a larger ssd for my laptop so I can start encoding more stuff directly instead of dumping it on the network.

Was that Tiger Lake encode fixed function or hybrid?? I am getting around 235 fps when encoding a blu ray to 1080p using fixed function. Using latest Intel graphics driver (31.0.101.3729). My encode settings for StaxRip are:-

--avhw --codec hevc --quality best --profile main10 --bframes 16 --b-pyramid --adapt-ltr --open-gop --async-depth 4 --no-repeat-pps --no-tskip --sao none --fixed-func --cqp 32:33:34 --qp-offset 2:4:8

[benwaggoner]

I'm pretty sure you can't have --bframes 16 for a Blu-ray compatible encode!

[RanmaCanada]

Was that Tiger Lake encode fixed function or hybrid?? I am getting around 235 fps when encoding a blu ray to 1080p using fixed function. Using latest Intel graphics driver (31.0.101.3729). My encode settings for StaxRip are:-

--avhw --codec hevc --quality best --profile main10 --bframes 16 --b-pyramid --adapt-ltr --open-gop --async-depth 4 --no-repeat-pps --no-tskip --sao none --fixed-func --cqp 32:33:34 --qp-offset 2:4:8

I used fixed function, and it was a 4k HDR movie, not 1080p.

[ukmark]

I'm pretty sure you can't have --bframes 16 for a Blu-ray compatible encode!

Not sure what this means. I ripped a blu ray disc to an mkv file to play back on my S905x3 box running CoreELEC. This plays back perfectly on my 4k TV. All my encodes are done this way. The source can be a blu ray disc or a higher bit rate x264 file, or my recorded TV shows etc, pretty much anything. I use StaxRip to encode. If I have a blu ray disc to encode, I decrypt first using MakeMKV and then encode the decrypted file.

All my encodes are kept on a 4TB HDD and played back on a laptop (via USB cable) or the S905x3 to a 4k TV (via HDMI cable). I don't play anything using a blu ray player. I have an external USB Blu ray drive connected to laptop, that I use (with MakeMKV) to decrypt and rip any blu ray discs I own.

Can u pls explain more? TIA

[ukmark]

I used fixed function, and it was a 4k HDR movie, not 1080p.

I would still have thought the speed would have been quicker. 4k is 4x size of blu ray, so a rough guess would have given between 50-60 fps (if I'm getting over 200 fps with 1080p). Maybe the HDR slows things down a lot. I have never ripped a 4k HDR file, so would not know. Still it's 10-15x faster than x265.

[ukmark]

Hybrid has a little more details in direct screenshot comparisons and the file size is a little smaller but it's not a big difference. Because of the big speed difference FF is the preferred choice in most cases. You can try without the custom offset by the way, maybe it's slightly better. CQP with 14 bframes is really good for sure. I tested with FFMPEG the last days, it's really good. But there is no SAO option to turn it off.

I have also been playing around with ICQ and 16 bframes using the latest Intel driver (31.0.101.3729) - (not sure if the graphics driver version makes a difference if you are using fixed-function encoding??) Admittedly a fairly quick test, but it does appear that ICQ and 16 bframes looks good (at least to my eyes). I also remember you saying that 14 bframes gave better results than 16 bframes and also with a little better compression. Again, using latest driver, it does appear that 16 bframes gives better compression than 14 bframes (CQP or ICQ). Just thought I'd pass it on - maybe worth a revisit? CQP still encodes quite a bit faster than ICQ using StaxRip and its output is so reliable.

[Yups]

I have also been playing around with ICQ and 16 bframes using the latest Intel driver (31.0.101.3729) - (not sure if the graphics driver version makes a difference if you are using fixed-function encoding??) Admittedly a fairly quick test, but it does appear that ICQ and 16 bframes looks good (at least to my eyes). I also remember you saying that 14 bframes gave better results than 16 bframes and also with a little better compression. Again, using latest driver, it does appear that 16 bframes gives better compression than 14 bframes (CQP or ICQ). Just thought I'd pass it on - maybe worth a revisit? CQP still encodes quite a bit faster than ICQ using StaxRip and its output is so reliable.


Do you make a PSNR/SSIM/VMAF comparison or how do you know 16 is better? Did you check the file size? bframes 14 file size is slightly smaller than bframes 16 in my testing, this is also an indication that bframes 14 is slightly more efficient. It could depend on the sample if this isn't the same for you. The difference is small anyways.

ICQ isn't bad but not as good as CQP. CQP with bframes 14-16 is extremely effective on Xe LP.

[benwaggoner]

I would still have thought the speed would have been quicker. 4k is 4x size of blu ray, so a rough guess would have given between 50-60 fps (if I'm getting over 200 fps with 1080p). Maybe the HDR slows things down a lot. I have never ripped a 4k HDR file, so would not know. Still it's 10-15x faster than x265.
HDR itself doesn't slow anything down. HDR requires 10-bit, and 10-bit is slower than 8-bit encoding, but that's Main versus Main10, not anything to do with color volume.

[benwaggoner]

Do you make a PSNR/SSIM/VMAF comparison or how do you know 16 is better? Did you check the file size? bframes 14 file size is slightly smaller than bframes 16 in my testing, this is also an indication that bframes 14 is slightly more efficient. It could depend on the sample if this isn't the same for you. The difference is small anyways.

ICQ isn't bad but not as good as CQP. CQP with bframes 14-16 is extremely effective on Xe LP.
Those are all single-frame metrics, so it's hard to interpret them with changes in B-frames. Harmonic mean of the prior two seconds isn't bad, but it can get complicated due to the different nature of reference and non-reference frames.

[Yups]

Those are all single-frame metrics, so it's hard to interpret them with changes in B-frames. Harmonic mean of the prior two seconds isn't bad, but it can get complicated due to the different nature of reference and non-reference frames.


The quality changes massively and you are saying they cannot interpret them, what? Subjective quality changes massively and of course the metric scores go up.

[ukmark]

Do you make a PSNR/SSIM/VMAF comparison or how do you know 16 is better? Did you check the file size? bframes 14 file size is slightly smaller than bframes 16 in my testing, this is also an indication that bframes 14 is slightly more efficient. It could depend on the sample if this isn't the same for you. The difference is small anyways.

ICQ isn't bad but not as good as CQP. CQP with bframes 14-16 is extremely effective on Xe LP.


Purely based on file size. So far 14 bframes gives a slightly bigger file size than 16 bframes (all other settings are the same). I don't compare PSNR/VMAF/SSIM, just an observation re file size and bframes. I've only tried a few files but each time, 14 bframes file size is bigger than 16 bframes with FF encoding. As you say, the difference is very small. I can't say that 14 bframes is better/worse quality than 16 bframes. To me the quality looks the same, but I have no metrics to qualify that, just that it appears to compress slightly better with 16 bframes and latest driver.

[ukmark]

Do you make a PSNR/SSIM/VMAF comparison or how do you know 16 is better? Did you check the file size? bframes 14 file size is slightly smaller than bframes 16 in my testing, this is also an indication that bframes 14 is slightly more efficient. It could depend on the sample if this isn't the same for you. The difference is small anyways.

ICQ isn't bad but not as good as CQP. CQP with bframes 14-16 is extremely effective on Xe LP.

Metrics for the first 10 minutes from my Star Trek into Darkness blu ray encode.

PSNR SSIM VMAF SPEED SIZE

14 bframes 36.22 0.9713 94.18 238 412 Mb
16 bframes 36.21 0.9711 94.13 239 403 Mb
Encode settings using StaxRip below:-

--avhw --codec hevc --quality best --profile main10 --bframes 14(or 16) --b-pyramid --adapt-ltr --open-gop --async-depth 4 --no-repeat-pps --no-tskip --sao none --fixed-func --cqp 30:32:33 --qp-offset 2:4:8

Differences are very small as can be seen. I have chosen to stick with 14 bframes, as even though the differences are small, I'd rather have that tiny bit extra quality and accept that file sizes could be around 2% bigger.

[benwaggoner]

The quality changes massively and you are saying they cannot interpret them, what? Subjective quality changes massively and of course the metric scores go up.
If you see massive subjective improvements, you don't need metrics! But sometimes subjective quality goes up and metrics go down, and vise versa.

For example, adaptive quantization in general tends to reduce metrics but (properly applied) improves subjective quality.

[Mister XY]

With all due respect to this effort and this work, using a score to orientate oneself brings 0 points. So you might have reduced a file to 200 MB, but especially in dark scenes you can see a lot of artefacts and blocks. So the only thing left to do is to check for yourself whether the settings have any effect or not. Example: Encoded the Intel film over and over again with different settings. In the x264 format, in the x265 (via software and hardware) format and also in the AV1 format. The funny thing was that my AV1 film and my x265 film were identical in size, but the AV1 film had a better score. Now comes the funny thing, the AV1 film had a very high level of blocking in dark scenes, even though it was better than the other formats according to the metrics. Therefore, these score values ​​have no meaning for me. The only benchmark that really exists is to take a film and encode it again and again, with different settings etc. and then you pay attention to fast transitions, to fine details in the film and to dark scenes and only when these are then in the encode still all are available, then you have reasonable settings. But using a score as a guide goes in the wrong direction. I can have a 100% score but not a viewable image because maybe the single image is 100% but not in conjunction with 23 other images that follow.

[rwill]

Yeah, single frame metrics combined with mean... questionable when there are multiple frames. Useless when the sequence contains more than one scene.

[GrandPa]

All you can do is buy it and try it, and report back.

The A380 card has arrived a few days ago, installation went without any problem, and I spent many hours with coding tests.

Summary
Configuration: i7-8700K on Z370 platform, GTX-1660 and A380 cards installed, BAR enabled, QSV HW decoding enabled

Basic preconditions for the coding tests: No visible blocks or artefacts in fast motion scenes or large dark, still areas
Procedure: Basis for comparison is a SW encode with x.265 at a reasonably high quality, VMAF score is taken as reference
(Absolute VMAF figures will not help here, just referring to the x.265 coding results)

Observations:
QSV is in most cases around 15-20% slower than NV (except with 4K material)
QSV is considerably faster than NV during downscaling (4K-to-720p, 1080p-to-720p)
QSV delivers better quality in most cases or better compression rate at same quality

StaxRip parameters:
--avhw --codec hevc --quality best --profile main10 --bframes 16 --b-pyramid --adapt-ltr --open-gop --async-depth 4
--colormatrix bt709 --colorprim bt709 --transfer bt709 --fixed-func --cqp ??:??:?? --qp-offset 2:4:8

A few sample results:

4K (Jellyfish) Speed [fps] File size [MB] VMAF
x.265 Reference 9 57,6 87,81
Nvenc 56 57,0 88,05
QSVenc 74 46,1 88,40

1080p (SolLevante) Speed [fps] File size [MB] VMAF
x.265 Reference 40 194,4 91,66
Nvenc 356 171,9 91,80
QSVenc 300 139,4 91,45

720p (HDTV Clip) Speed [fps] File size [MB] VMAF
x.265 Reference 136 23,9 89,33
Nvenc 855 23,0 89,13
QSVenc 629 17,9 89,89
(No audio track contained, CQP parameters tuned accordingly aiming at the reference quality)

Overall, the A380 card is a good deal, although there seems to be still room left for optimization, maybe with future
driver releases.

[ukmark]

I love StaxRip but it appears that development has stopped. It is easily the best gui I have used and does everything very well. I'm mainly interested in the gpu hardware quick sync side of things, rather than the software/CPU encoding.

Does anybody know of any other current encoder gui that will incorporate up-to-date support of quick sync and x265 etc along with audio support etc??

I know Handbrake is still going, but you are limited in the advanced options that can be set (unless things have changed). For example the "qp-offsets" can be set in StaxRip and these really help in maintaining quality whilst keeping file size as low as possible. This option is not available in Handbrake. This is not the basic QP I,P,B offsets but the
offsets under "rate control" section of the StaxRip gui. It may well be that even in its current state StaxRip will suffice for years to come as regards HEVC quick sync support, but wanted to know if there was anything else out there with the same advanced features etc.

I have looked on the web but could not find anything. Free or paid does not matter, but would prefer free.

TIA

[Yups]

Metrics for the first 10 minutes from my Star Trek into Darkness blu ray encode.

PSNR SSIM VMAF SPEED SIZE

14 bframes 36.22 0.9713 94.18 238 412 Mb
16 bframes 36.21 0.9711 94.13 239 403 Mb
Encode settings using StaxRip below:-

--avhw --codec hevc --quality best --profile main10 --bframes 14(or 16) --b-pyramid --adapt-ltr --open-gop --async-depth 4 --no-repeat-pps --no-tskip --sao none --fixed-func --cqp 30:32:33 --qp-offset 2:4:8

Differences are very small as can be seen. I have chosen to stick with 14 bframes, as even though the differences are small, I'd rather have that tiny bit extra quality and accept that file sizes could be around 2% bigger.


In all my tests bframes 14 file size was slightly smaller and scores a bit higher, for example in the Intel video. In your test there is not much difference anyways. 14-16 are equally good overall.

Did you try without custom offset with like 30:32:36? I think disabling tskip might not be better in every case.

You don't need LTR and repeat pps with CQP, it makes no difference.


I love StaxRip but it appears that development has stopped. It is easily the best gui I have used and does everything very well. I'm mainly interested in the gpu hardware quick sync side of things, rather than the software/CPU encoding.

Does anybody know of any other current encoder gui that will incorporate up-to-date support of quick sync and x265 etc along with audio support etc??



Maybe have a look to Hybrid (https://forum.doom9.org/showthread.php?t=153035)

It's not up to date (no AV1 support yet) but it's not stopped like Staxrip.

[ukmark]

Maybe have a look to Hybrid (https://forum.doom9.org/showthread.php?t=153035)

It's not up to date (no AV1 support yet) but it's not stopped like Staxrip.

Cheers.

[RanmaCanada]

I love StaxRip but it appears that development has stopped. It is easily the best gui I have used and does everything very well. I'm mainly interested in the gpu hardware quick sync side of things, rather than the software/CPU encoding.

Does anybody know of any other current encoder gui that will incorporate up-to-date support of quick sync and x265 etc along with audio support etc??

I know Handbrake is still going, but you are limited in the advanced options that can be set (unless things have changed). For example the "qp-offsets" can be set in StaxRip and these really help in maintaining quality whilst keeping file size as low as possible. This option is not available in Handbrake. This is not the basic QP I,P,B offsets but the
offsets under "rate control" section of the StaxRip gui. It may well be that even in its current state StaxRip will suffice for years to come as regards HEVC quick sync support, but wanted to know if there was anything else out there with the same advanced features etc.

I have looked on the web but could not find anything. Free or paid does not matter, but would prefer free.

TIA

You can always update the encoder files (https://github.com/rigaya/QSVEnc) manually in staxrip and then it will ask you what the new version fo the software is. Best way to stick with a gui you like and know how to use.

[ukmark]

Thanks for that - yes I do update QSVEnc regularly and am using the latest version (v7.23). Maybe stax76 will return and continue with StaxRip in the future. It's definitely the best GUI I have used so far.

[ReinerSchweinlin]

So my ARC 770 arrived :)

Where do I start? From reading this thread, I remember that with Intel XE, the QSVENCC from rigaya, called upon by staxrip, was the best solution for quality. The ffmpeg implementation did have some problems? A´S Video Encoder seems to beusing the intel provided encoder, which offers a lot less knobs to turn.... Handbrake also offers only the bare minimum to tune.

SO using hybrid/staxrip or any other QSVENCC GUI is still the best option? (for the goal of getting max quality out of encodes..)

[Yups]

QSVEnc, FFmpeg and Handbrake. FFmpeg HEVC works fine, the only missing option for me is SAO. I think QSVEnc GUI is the best to begin with.

[ReinerSchweinlin]

Thanx, so I´ll probably use hybrid as GUI for QSVENC :)

[GrandPa]

Overall, the A380 card is a good deal, although there seems to be still room left for optimization, maybe with future
driver releases.

Indeed, with the latest driver release 3959, coding speed has been considerably improved in several test cases. I'll continue monitoring the promising evolution.

[Lan4]

Hello! Help me please. What is VQEnhancer? I can't find a description for this.

[DuyNghia.Net]

Hello everyone, I am about to buy Intel UHD 770 12th/13th Gen for encoding HEVC QSV Encoding 1080p/4K.

How fast is it compared to A380? Anyone, Thanks!

My 9th gen Intel 9400 UHD630 gets Constant 60fps 1080p transcoding H264->HEVC using QSVEnc which is quite slow!

[RanmaCanada]

Hello everyone, I am about to buy Intel UHD 770 12th/13th Gen for encoding HEVC QSV Encoding 1080p/4K.

How fast is it compared to A380? Anyone, Thanks!

My 9th gen Intel 9400 UHD630 gets Constant 60fps 1080p transcoding H264->HEVC using QSVEnc which is quite slow!

They should be about the same as they have the same ASICS as far as I understand. The dedicated graphics might be a bit faster due to cpu overhead. If I am mistaken, can someome please correct me as I do not want to be spreading lies.

[Yups]

Hello everyone, I am about to buy Intel UHD 770 12th/13th Gen for encoding HEVC QSV Encoding 1080p/4K.

How fast is it compared to A380? Anyone, Thanks!

My 9th gen Intel 9400 UHD630 gets Constant 60fps 1080p transcoding H264->HEVC using QSVEnc which is quite slow!



HD630 features a hybrid only HEVC encoder which is slow. UHD 770 can do both hybrid and fully fixed encoding. UHD 770 and A380 are much faster and better quality. Note that A380 als supports AV1 encoding.

You can see here:

https://forum.doom9.org/showpost.php?p=1937452&postcount=396

Iris Xe has the same encoder as UHD 770.

[easyfab]

New intel lunar lake with Xe2 is out now.

Is better encoding with HEVC or AV1 expected ?

[RanmaCanada]

New intel lunar lake with Xe2 is out now.

Is better encoding with HEVC or AV1 expected ?

The current wikipedia (https://en.wikipedia.org/wiki/Intel_Quick_Sync_Video) page only goes up to Arrow Lake (who's page doesn't exist), so uncertain. It would be great, but with everything that is exploding at Intel lately, more than likely not.

Edit: since I can't strike this through, I'll add this pdf (https://cdrdv2-public.intel.com/824434/2024_Intel_Tech%20Tour%20TW_Xe2%20and%20Lunar%20Lakes%20GPU.pdf) from Intel. Apparently the new Lunar Lake has hardware VVC encoding, which would potentially mean new ASICS and possibly better encoding for everything else. Until someone buys one, and tests it, we won't know for sure.

[ReinerSchweinlin]

Seems VCC Decode is on board, I see no clear mentioning of encoding via hardware.

[nevcairiel]

Lunar Lake has VVC Decode, not Encode. Arrow Lake is expected to get 1st Gen Xe graphics only, so no VVC there.