Repository navigation
Releases: 5fish/SVT-AV1
Release list
5fish/SVT-AV1 v2.3.261021
Changes
- Better DLF default at high
--crf.
Build Information
Ref refs/heads/main | Commit ad31e0ef10f872ccf731da9b7f1a14197c3d57bc | Dolby Vision & HDR10+
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 + znver2 | 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 + znver2 | LTO PGO
macOS | arm64 | LLVM Project Clang LLD | armv8.5-a+simd+crypto + apple-m3 | 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 + znver5 ⇒ znver2 ⇒ x86-64-v3 + znver2.
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 + znver2 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, Khaoklong, Miss Ashenlight, Soda, SwareJonge, Trix, Yiss, and other people in the community for discovering the method to build the fastest binary.
5fish/SVT-AV1 v2.3.261005
Changes
- Fixed an error in documentation and encoder error message.
Build Information
Ref refs/heads/main | Commit 1170a8d58a9a6855fa7fe88359f091dcfeb9e434 | Dolby Vision & HDR10+
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 + znver2 | 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 + znver2 | LTO PGO
macOS | arm64 | LLVM Project Clang LLD | armv8.5-a+simd+crypto + apple-m3 | 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 + znver5 ⇒ znver2 ⇒ x86-64-v3 + znver2.
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 + znver2 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, Khaoklong, Miss Ashenlight, Soda, SwareJonge, Trix, Yiss, and other people in the community for discovering the method to build the fastest binary.
5fish/SVT-AV1 v2.3.261004
Changes
- Tested out the default of
--balancing-mg-dist-q-biasand--balancing-noise-level-q-bias. The new features are now ready for production. --psy-bias-dgand new--balancing-q-biaslogic improves bitrate allocation.
The quality difference between the worst frames and the best frames are narrowed, and the encode should be more consistent.
The new features are enabled by default by the two base-psy-bias. No additional parameters need to be set. However, if you aim for even more consistency than the default we use, try--balancing-mg-dist-q-bias 4.
Build Information
Ref refs/heads/main | Commit 6eb60b04b54b220069817124ac6cae2b7c8e73d6 | Dolby Vision & HDR10+
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 + znver2 | 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 + znver2 | LTO PGO
macOS | arm64 | LLVM Project Clang LLD | armv8.5-a+simd+crypto + apple-m3 | 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 + znver5 ⇒ znver2 ⇒ x86-64-v3 + znver2.
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 + znver2 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, Khaoklong, Miss Ashenlight, Soda, SwareJonge, Trix, Yiss, and other people in the community for discovering the method to build the fastest binary.
5fish/SVT-AV1 v2.3.260929
Changes
- Revert cost based early exit in introduced in v2.3.260827 thanks to Del.
Testing
This version is currently in testing, which means although we've seen improvements on the sources we've tested, we're not yet sure that we'll see improvements and not degradations on different kinds of sources. Use with caution.
Build Information
Ref refs/heads/psy_bias_dg | Commit 5bbf90a44bf25a6aebdf30c3bb9f28c8c102f1ff | Dolby Vision & HDR10+
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 + znver2 | 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 + znver2 | LTO PGO
macOS | arm64 | LLVM Project Clang LLD | armv8.5-a+simd+crypto + apple-m3 | 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 + znver5 ⇒ znver2 ⇒ x86-64-v3 + znver2.
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 + znver2 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, Khaoklong, Miss Ashenlight, Soda, SwareJonge, Trix, Yiss, and other people in the community for discovering the method to build the fastest binary.
5fish/SVT-AV1 v2.3.260829
Changes
- Implement
--psy-bias-dg. - Implement
--balancing-mg-dist-q-biasand revamp--balancing-noise-level-q-bias(additional tuning required).
Testing
This version is currently in testing, which means although we've seen improvements on the sources we've tested, we're not yet sure that we'll see improvements and not degradations on different kinds of sources. Use with caution.
Build Information
Ref refs/heads/psy_bias_dg | Commit 7f847c6b4b6115cd0454f95a4494f87d5acca8a4 | Dolby Vision & HDR10+
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 + znver2 | 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 + znver2 | 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 + znver5 ⇒ znver2 ⇒ x86-64-v3 + znver2.
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 + znver2 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, Khaoklong, Miss Ashenlight, Soda, SwareJonge, Trix, Yiss, and other people in the community for discovering the method to build the fastest binary.
5fish/SVT-AV1 v2.3.260828
Changes
- Early version of
--psy-bias-dg(additional tuning required).
Testing
This version is currently in testing, which means although we've seen improvements on the sources we've tested, we're not yet sure that we'll see improvements and not degradations on different kinds of sources. Use with caution.
Build Information
Ref refs/heads/psy_bias_dg | Commit 49577636691220f21546417976660b5a76843d33 | Dolby Vision & HDR10+
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 + znver2 | 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 + znver2 | 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 + znver5 ⇒ znver2 ⇒ x86-64-v3 + znver2.
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 + znver2 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, Khaoklong, Miss Ashenlight, Soda, SwareJonge, Trix, Yiss, and other people in the community for discovering the method to build the fastest binary.
5fish/SVT-AV1 v2.3.260827
Changes
- Cost based early exit via
--psy-bias-mds0-sad. --psy-bias-mds0-sadis now enabled by default at--lineart-psy-bias [>= 2]or--texture-psy-bias [>= 3].- Adopt further ssim rdmult scaling tuning from Clybius/svt-av1-hdr-personalized.
Build Information
Ref refs/heads/main | Commit 1c4fb91a06b0b14aafefe11835dafd91d337a3a4 | Dolby Vision & HDR10+
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 + znver2 | 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 + znver2 | 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 + znver5 ⇒ znver2 ⇒ x86-64-v3 + znver2.
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 + znver2 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, Khaoklong, Miss Ashenlight, Soda, SwareJonge, Trix, Yiss, and other people in the community for discovering the method to build the fastest binary.
5fish/SVT-AV1 v2.3.260818
Changes
- Enable high DLF strategy at
--crf [> 28.00]by default. - Adopt a TPL tuning from
--tune 3to--balancing-q-bias. - Adopt ssim rdmult scaling tuning from Clybius/svt-av1-hdr-personalized.
Testing
This version is currently in testing, which means although we've seen improvements on the sources we've tested, we're not yet sure that we'll see improvements and not degradations on different kinds of sources. Use with caution.
Build Information
Ref refs/heads/psy-bias-adjust | Commit 825aaa4121c49d093b3b550a376ee407fdd4ce1e | Dolby Vision & HDR10+
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 + znver2 | 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 + znver2 | 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 + znver5 ⇒ znver2 ⇒ x86-64-v3 + znver2.
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 + znver2 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, Khaoklong, Miss Ashenlight, Soda, SwareJonge, Trix, Yiss, and other people in the community for discovering the method to build the fastest binary.
5fish/SVT-AV1 v2.3.260719
Changes
- Adopted NEON intrinsics for QM from -HDR.
- Adjusted
--noise-level-thrdefault on--crf [> 30.00].
Build Information
Ref refs/heads/main | Commit 45bb53193b66b481fe307f51cfcb67e8c02d041e | Dolby Vision & HDR10+
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 + znver2 | 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 + znver2 | 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 + znver5 ⇒ znver2 ⇒ x86-64-v3 + znver2.
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 + znver2 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, Khaoklong, Miss Ashenlight, Soda, SwareJonge, Trix, Yiss, and other people in the community for discovering the method to build the fastest binary.
5fish/SVT-AV1 v2.3.260707
Changes
- Upgraded to a new building system. Now providing dovi and hdr10plus. No functional changes in this version.
Build Information
Ref refs/heads/main | Commit acb42928f59ebb0313ea39ca614389b6e6d8a9b7 | Dolby Vision & HDR10+
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 + znver2 | 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 + znver2 | 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 + znver5 ⇒ znver2 ⇒ x86-64-v3 + znver2.
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 + znver2 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, Khaoklong, Miss Ashenlight, Soda, SwareJonge, Trix, Yiss, and other people in the community for discovering the method to build the fastest binary.