Trying out the new AV1 profiles and finding wondering if there might be some tuning or something or if it’s just that hardware acceleration isn’t working well for me.
Some history, and maybe this will explain my confusion - I typically use MP4 HEVC profile and a couple months back got a new pc with an intel arc video card so I’d been trying out the hardware encoding. What I found was my end result mp4 (from a TS original) was maybe 10-20% larger in size (the output mp4’s are probably 1/5 to 1/10 size of original ts) with hardware acceleration on so at first glance I’d go with hardware encoding off, however the trade off here is speed where time to convert was maybe 1/10 with hardware acceleration on
So with hardware acceleration file size is a bit worse, but processing time is much better.
Now with AV1 I’m seeing a difference in behavior.
A 5.5GB TS file converted:
with H/A to 88 mb in just under 7 minutes.
without H/A to 158 mb in about 48 minutes.
So for AV1 my processing with H/A is faster and smaller (whereas HEVC it’s faster but a bit bigger) however the caveat here is in HEVC the video quality in comparable between both conversions, whereas with AV1 the processing with H/A has a much poorer look to it.
The file sizes of the AV1 are definitely better. I only have the H/A HEVC numbers but same input file was output as 650mb in 7 minutes of processing and I think that one is reasonably comparable video quality of the AV1 without H/A, I just don’t really want the longer processing times.
Really just posting this as I’m wondering what other peoples experiences are. I do also have a machine with an Nvidia card that can do H/A so I might try it on the weekend to see how the AV1 is there.
Can you attach your conversions logs for with and without hardware acceleration for the AV1 files. There’s an algorithm in play here which may be causing the difference and logs will show what’s going on.
logs.zip (1.8 MB)
I’ve attached the 2 AV1 conversion logs and I had also just done an HEVC one so threw it in as well. I renamed them as you’ll see just to make it easier for me to remember which was which.
Okay, made some adjustments. Try it again (preferably the same video and settings) with todays 2.6.3 beta version (you don’t need to uninstall 2.6.2, you can directly upgrade to 2.6.3 beta) and share the logs.
FYR, can’t really compare the AV1 with HEVC because the HEVC is about 350kbps vs 1.3mpbs for HEVC. So if the visual quality is the same, that’s the power of AV1 in the compression that you’re seeing.
Also if you get a chance can you make one edit to the MP4 AV1 profile you’re using and change:
order=ffmpeg and then re-run the same file with and without HA and attach those also. It will help us compare and optimize the quality/performance across different encoders to provide a more consistent experience.
v263_conversionLogs.zip (2.4 MB)
Attached the 4 logs. The runs with hardware acceleration are both visually quite a bit worse than the non-hardware accelerated.
I was going to also try a conversion on Nvidia but it’s an older card and doesn’t support AV1, so think it falls back one of the other profiles so only got a software conversion.
Thanks for the logs, very helpful. Made a few adjustments (and a bug fix) to the AV1 algorithm. Try today’s 2.6.3 beta. You should see an improvement in visual quality as well as a more consistent quality between the HA and non HA profiles.
Can you re-run your tests and attach the 4 logs like before for the handbrake and the ffmpeg profiles so I can see the results from your machine.
beta3_20240108_Logs.zip (2.4 MB)
attached. I’ll take a closer look later but the video quality is definitely better.
Thanks. Made a slight tweak in today’s 2.6.3 beta which should bring all 4 about inline in terms of quality/bitrate and maybe a slight improvement in the software encoding speed.
Ok I’ll give it a try but might not be till later this week.