Dav1d 0.6 AV1 Video Decoder Delivering Healthy Improvements For Intel + AMD Processors

Given this week's release of dav1d 0.6, here are some fresh benchmarks of this open-source AV1 video decoder on a few different Intel and AMD systems so far.

Dav1d 0.6 has various AVX2 and AVX-512 optimizations that excite us plus 10/12-bit video decoding improvements for ARMv8 hardware. With waiting for some more exciting ARM server platforms still to arrive, this round of testing is just looking at the Intel/AMD x86_64 CPU performance.

Several different higher-end Intel and AMD systems were benchmarked for the initial dav1d 0.6.0 benchmarking. Further system tests will be coming up in the days ahead. Note that there are software/hardware differences between systems so these benchmarks aren't about comparing the CPU performance directly but more about just looking at the dav1d 0.6 performance on several different systems.

Even with 1080p content using the "Chimera" sample, both Intel and AMD were seeing some modest improvements on this new release.

At 4K, dav1d 0.6.0 was still providing a measurable advantage over v0.5.0.

But where things really got interesting for the Intel/AMD processors was the much faster 10-bit decoding of AV1 content with dav1d 0.6.0 for both Intel and AMD. Very significant gains in this case.

While more dav1d 0.6.0 benchmarks are being worked on, the current ones in full along with all of the system details, etc, can be found via this OpenBenchmarking.org result file.