Intel Engineers Release SVT-AV1 1.3 With More Optimizations, Tuning For Better AV1 Encode

[phoronix]

Phoronix: Intel Engineers Release SVT-AV1 1.3 With More Optimizations, Tuning For Better AV1 Encode

The open-source SVT-AV1 encoder for AV1 content that is primarily developed by Intel software engineers continues getting faster and finding new ways to provide better tuned presets and yield better performance for AV1 encoding on CPUs...

https://www.phoronix.com/news/SVT-AV1-1.3

[birdie]

To fuck with patents, vvenc encodes a ton faster and a ton better (in terms of preserving fine details).

[Quackdoc]

svtav1 is nice, it can do real time encoding (being more then 30fps and often 60fps) at 1080p at decent qualities, a shame it only supports 4:2:0 however

[zxy_thf]

To fuck with patents, vvenc encodes a ton faster and a ton better (in terms of preserving fine details).

To be fair, SVT-AV1 is far from being a competitive encoder: https://www.compression.ru/video/cod...4k_report.html
It's even worse than some HEVC encoders.

[Quackdoc]

To be fair, SVT-AV1 is far from being a competitive encoder: https://www.compression.ru/video/cod...4k_report.html
It's even worse than some HEVC encoders.

you probably shouldn't judge by these, they are often flawed and extremely biased, it's easy enough to tell, SVTAV1 is the lowest fidelity of the bunch, but it's typically faster AND smaller then HEVC at a given size until you hit svtav1's (or really all of the encoders right now) quality ceiling, but that ceiling is typically higher then most people will care.

[Chugworth]

I currently use the command below for general-purpose AV1 video encoding:

ffmpeg -i infile.mp4 -c:v libsvtav1 -qp 35 -preset 4 -c:a libopus -b:a 128k outfile.mkv

It does go fairly slow, and if you want it to go MUCH faster you could change "-preset 4" to "-preset 8". However, if you take two video files and ALT+TAB between a single frame in them, it is possible to see a subtle loss of detail between 4 and 8. So I generally settle on 4 just for the slightly better quality.

Depending on the video though, it is possible to get some massive space savings and sometimes shave several gigs off of a single video file with no real noticeable loss in detail.

[Quackdoc]

I currently use the command below for general-purpose AV1 video encoding:

ffmpeg -i infile.mp4 -c:v libsvtav1 -qp 35 -preset 4 -c:a libopus -b:a 128k outfile.mkv

It does go fairly slow, and if you want it to go MUCH faster you could change "-preset 4" to "-preset 8". However, if you take two video files and ALT+TAB between a single frame in them, it is possible to see a subtle loss of detail between 4 and 8. So I generally settle on 4 just for the slightly better quality.

Depending on the video though, it is possible to get some massive space savings and sometimes shave several gigs off of a single video file with no real noticeable loss in detail.

I typically use presets 6, 8, and 11, depending on what im doing, it's also worth noting that svtav1 supports crf so for a lot of folk that might be the better option, but other than that I think most people would be satisfied with this.

ofc when possible I just use av1an with aom-av1-psy

[Chugworth]

I typically use presets 6, 8, and 11, depending on what im doing, it's also worth noting that svtav1 supports crf so for a lot of folk that might be the better option, but other than that I think most people would be satisfied with this.

ofc when possible I just use av1an with aom-av1-psy

Ah, good to know. I think that not too long ago the -crf switch wouldn't work with libsvtav1, which is why I was using -qp. It looks like it does work now, so it would probably be better to use -crf 35 rather than -qp 35. I'll have to do some testing and see how it compares.

[wswartzendruber]

Can it emulate film grain yet, or does it still just denoise and not perform noise modeling?

[Chugworth]

Can it emulate film grain yet, or does it still just denoise and not perform noise modeling?

From the ffmpeg AV1 page:

SVT-AV1 supports film grain synthesis, an AV1 feature for preserving the look of grainy video while spending very little bitrate to do so. The grain is removed from the image with denoising, its look is approximated and synthesized, and then added on top of the video at decode-time as a filter.

The film grain synthesis feature is invoked with -svtav1-params film-grain=X, where X is an integer from 1 to 50. Higher numbers correspond to higher levels of denoising for the grain synthesis process and thus a higher amount of grain.

I haven't tested that feature myself though.