AV1
- JohnAStebbins
- HandBrake Team
- Posts: 5723
- Joined: Sat Feb 09, 2008 7:21 pm
Re: AV1
Yes, we had a somewhat lively discussion in the AV1 issue thread on github yesterday https://github.com/HandBrake/HandBrake/issues/457
TL;DR, AV1 is about 30% better than x264 in quality vs. file size metrics. And it's about 6000x slower. I.e. you can make your files 30% smaller if you are willing to wait 9000 hours (a.k.a. 1 year) to encoder your feature length HD movie.
From the Facebook article:
TL;DR, AV1 is about 30% better than x264 in quality vs. file size metrics. And it's about 6000x slower. I.e. you can make your files 30% smaller if you are willing to wait 9000 hours (a.k.a. 1 year) to encoder your feature length HD movie.
From the Facebook article:
On the other hand, the encoding computational complexity (in terms of encoding run time) of AV1 compared with x264 main, x264 high and libvpx-vp9 for CRF/QP mode was increased by factors of 5721.5x, 5869.9x and 658.5x, respectively, as shown in Figure 4.
-
- Veteran User
- Posts: 4854
- Joined: Wed May 04, 2011 11:06 pm
Re: AV1
I did see they mentioned 10 second clips that they tested it on and my napkin math put that at 12+ hours to encode
I guess if you have tons of CPUs and are serving up billions of Taylor Swift videos it might make sense to save 30%.
I guess if you have tons of CPUs and are serving up billions of Taylor Swift videos it might make sense to save 30%.
- JohnAStebbins
- HandBrake Team
- Posts: 5723
- Joined: Sat Feb 09, 2008 7:21 pm
Re: AV1
When VP9 was initially pushed out the door, it was also extremely slow (but still faster than current AV1 by about 12x). It took a bit over 2 years for them to bring VP9 encode speed up to the point that it was acceptable for inclusion in HandBrake. I expect AV1 will be on a similar trajectory and will be keeping an eye on it as things develop. In that time, CPU speeds will increase and perhaps we will see dedicated hardware encoders supporting AV1 built into CPU or GPU silicon.rollin_eng wrote: ↑Thu Apr 12, 2018 4:15 pm I guess if you have tons of CPUs and are serving up billions of Taylor Swift videos it might make sense to save 30%.
Please don't t take my comments as any disparagement against AV1. I would really like to see it succeed. But some patience is in order here.
-
- Veteran User
- Posts: 4854
- Joined: Wed May 04, 2011 11:06 pm
Re: AV1
Oh I’m not advocating for its inclusion in HB, just putting this out there for discussion.
I also see that Apple and Samsung are not involved as I imagine a lot of videos for YouTube and Facebook come from their devices it will be interesting to see what direction they go.
I also see that Apple and Samsung are not involved as I imagine a lot of videos for YouTube and Facebook come from their devices it will be interesting to see what direction they go.
- JohnAStebbins
- HandBrake Team
- Posts: 5723
- Joined: Sat Feb 09, 2008 7:21 pm
Re: AV1
For those interested in the old VP9 saga:
viewtopic.php?f=26&t=27343&hilit=month
That thread explains why I am conservative about my estimates for an AV1 HandBrake roadmap.
Starting in June 2013, I did regular benchmarks as new release came out and posted my results to the forums. When VP9 reached an acceptable state (May 2016), we added support for it to HandBrake. I'll do the same with AV1, but post results to the github issue thread instead since that's were we track development status and issues now.
viewtopic.php?f=26&t=27343&hilit=month
That thread explains why I am conservative about my estimates for an AV1 HandBrake roadmap.
Starting in June 2013, I did regular benchmarks as new release came out and posted my results to the forums. When VP9 reached an acceptable state (May 2016), we added support for it to HandBrake. I'll do the same with AV1, but post results to the github issue thread instead since that's were we track development status and issues now.
- JohnAStebbins
- HandBrake Team
- Posts: 5723
- Joined: Sat Feb 09, 2008 7:21 pm
Re: AV1
LOL. My 1 year estimate is for a 90 minute HD movie. So 180 minute UHD is x2 for duration and x4 for resolution, resulting in 8 years encode time. So yes, by then I would fully expect hardware and software improvements would make the next encode much faster