AOM AV1 v3.4 Encoder Brings Better Performance

[phoronix]

Phoronix: AOM AV1 v3.4 Encoder Brings Better Performance

Google engineers on Friday released AOM AV1 v3.4 as the newest version of this open-source AV1 CPU-based video encoder...

https://www.phoronix.com/scan.php?page=news_item&px=AOM-AV1-3.4-Released

[archsway]

How does this compare to earlier versions?

What was performance like a year ago? Two years ago?

[Michael]

How does this compare to earlier versions?

What was performance like a year ago? Two years ago?

Drop down menu "Analyze Test Configuration" on https://openbenchmarking.org/test/pts/aom-av1 and select older version(s).

[FPScholten]

How does this compare to earlier versions?

What was performance like a year ago? Two years ago?

I haven been compiling and using AOM-av1 for about 2 years now and have seen a gradual improvement in speed. Better yet, the (visual) quality did not degrade over that time, as faster encoding usually leads to lesser quality. When using the right settings AOM-AV1 encodes almost as fast as SVT-AV1 but with better quality and smaller files.

[linuxgeex]

How does this compare to earlier versions?

What was performance like a year ago? Two years ago?

about 20% faster than last year (2.x-3.0), and about 600% faster than 2 years ago (1.x)

All of last year libaom-av1 was very competitive with svt-av1, to no surprise, because google merged them.

-cpu-used 4 is competitive with x265 veryslow, being slightly faster, slightly better quality at the same bitrate, with a caveat that it achieves markedly fewer artefacts in moderate motion areas while being slightly less sharp when bitrate-starved, ie ~600kbps for 1080p.

I should echo @FPScholten's sentiment that these speed improvements came without quality degradation, and if anything, the -cpu-used options 6 and lower (slower speeds) have shown quality improvements enabled by the speed improvements, which allowed the developers to enable more "experiments" in the same time budget.