Yardsticks

A speed figure compared only with our own build says little. Each speed claim on these pages is checked against an independent implementation, named here, and the gaps that check found are listed in the full write-up, worst first.

Which implementation each claim is measured against

ClaimYardstick
Native encode speed per preset, Arm and x86x264, x265, SVT-AV1, libaom, libvpx and openh264 at their real-time settings, timed in the same rounds; NVENC for scale
H.264 and HEVC decodeFFmpeg's decoders
AV1 decodedav1d (VideoLAN)
VP9 decodevpxdec (libvpx) and FFmpeg's VP9 decoder
VVC decodeVVdeC (Fraunhofer HHI)
AV2 decodeavmdec, the research reference decoder (not built for speed)
WebAssembly decode and encodethe official single-thread ffmpeg.wasm (a weak yardstick: a generic build without hand-written SIMD, and no AV1 or VVC decoder)
Colour conversion and the demo's display pathlibyuv
Time to the first packet, memorythe same reference encoders and decoders, measured the same way

Rules of the comparison

Read the measurements

We thank the FFmpeg, dav1d, libvpx, VVdeC, AVM, libyuv, x264, x265, SVT-AV1, libaom and openh264 projects: their implementations are the bar these numbers are held to.