Skip to content

5fish/SVT-AV1 v2.3.260627

Choose a tag to compare

@github-actions github-actions released this 28 Jun 08:20

Changes

  • Scene Change Detection is now enabled by default.
    • Not setting anything enables internal scene change detection with an optimal config for --balancing-q-bias system used by --lineart-psy-bias / --texture-psy-bias.
    • Setting --keyint -1 or --keyint 0 --scd 0 disables scene change detection. If you're using an encoding framework like av1an or vs-muxtools, then they have already set that for you.
  • This version fixed an inter-GOP bug regarding cand elimination.
  • This version introduced Yiss' fix on --enable-tf 2.

Build Information

Ref refs/heads/main | Commit 80fc61e6dcb9cfc7f8dd080117db310b732040eb
Windows | x86-64 | LLVM Project Clang LLD | winthread | icelake-server + znver5 | LTO PGO
Windows | x86-64 | LLVM Project Clang LLD | winthread | znver2 | LTO PGO
Windows | x86-64 | LLVM Project Clang LLD | winthread | x86-64-v3 AVX2 Max | LTO PGO
Windows | arm64 | LLVM Project Clang LLD | winthread | armv8.7-a+crypto+sm4+sha3+fp16+sve+sve2 + oryon-1 | LTO PGO
Linux | x86-64 | LLVM Project Clang LLD | icelake-server + znver5 | LTO PGO
Linux | x86-64 | LLVM Project Clang LLD | znver2 | LTO PGO
Linux | x86-64 | LLVM Project Clang LLD | x86-64-v3 AVX2 Max | LTO PGO
macOS | arm64 | LLVM Project Clang LLD | armv8.5-a+simd+crypto + apple-m3 | LTO PGO
macOS | x86-64 | LLVM Project Clang LLD | skylake | LTO PGO

On x86-64 systems, no matter if you're on an Intel or AMD system, you should try these builds in the order of icelake-server + znver5znver2x86-64-v3.
If your CPU doesn't support the features required by the icelake-server + znver5 version, often indicated by crashing outright when attempting to make any encodes, you can try the znver2 version and see if that works. If not, then the x86-64-v3 version, which should always work on any modern systems.
Some versions might not be available if the GitHub Actions runners we get this time don't support the respective features.

Thanks to studies and contributions from afed, Emre, Ironclad, Itachi Uchiha, Miss Ashenlight, Soda, SwareJonge, Trix, Yiss and other people in the community for discovering the method to build the fastest binary. If you need to build a binary yourself, check the note in the building script.