Intel SVT-AV1 0.8 AV1 Video Encoding Benchmarks
On Friday Intel released SVT-AV1 0.8 with more AVX2/AVX-512 optimizations for this one of the fastest CPU-based AV1 open-source video encoders (and growing decoding support too). Here are some benchmarks of SVT-AV1 0.8 compared to the previous v0.7 release on various Intel and AMD systems.
Over the weekend I started tossing SVT-AV1 0.8 on various systems via the Phoronix Test Suite / OpenBenchmarking.org for seeing how the performance compares to the previous Scalable Video Technology AV1 encoder release. The Ubuntu Linux systems picked spanned various generations but mostly a random assortment of hardware based upon convenience for this one-page weekend testing. The systems included the Intel Core i7 1065G7, Core i7 7740X, Core i9 9900KS, Xeon E5-2687W v3, dual Xeon Gold 6138, dual Xeon Platinum 8280 on the Intel side. On the AMD side was the Ryzen 5 3600X, Threadripper 3960X, EPYC 7601, and dual EPYC 7742.
With encoding mode "0" for the highest-quality AV1 encodes, the SVT-AV1 0.8 performance was indeed faster than the previous release on all of the CPUs tested.
With the more reasonable encoding mode four for a fair trade-off of speed and encode quality, the Intel CPUs were seeing better performance out of the new release but this did hit the AMD CPU performance. Though even with this hit to the performance, the dual EPYC 7742 did still outperform the dual Xeon Platinum 8280, for reference.
At the encode mode quality of "8" for faster encode performance, all of the systems saw slower performance -- in some cases by quite significant margins. In fact, quite surprising regressions, so presumably some of their optimizations still need to be tuned better before seeing SVT-AV1 1.0.
See how your Linux system(s) compare by installing the Phoronix Test Suite and running phoronix-test-suite benchmark svt-av1.