The release of the SVT-AV1 3.0 library (Scalable Video Technology AV1) has been published, featuring implementations of both encoder and decoder for the AV1 video encoding format, with hardware parallel computation capabilities present in modern Intel CPUs being utilized for acceleration. The project was created by Intel in partnership with Netflix with the aim of achieving a performance level suitable for real-time video transcoding and applications in video-on-demand (VOD) services. Currently, development is conducted under the aegis of the Open Media Alliance (AOMedia), which oversees the development of the AV1 video encoding format. Previously, the project was developed within the OpenVisualCloud initiative, which also develops the SVT-HEVC and SVT-VP9 encoders. The code is distributed under the BSD license.
SVT-AV1 can be built for systems based on any architecture for which a C99-compliant compiler exists, but the best performance is achieved on x86_64 systems where assembly optimizations based on SIMD instructions are applied (AVX2 support in the CPU is preferable, but SS2 support is sufficient as a minimum). Memory consumption depends on the number of CPU cores used in encoding, regulated by the option "—lp". Due to the complexity of the algorithms used in AV1, encoding this format requires significantly more resources than other formats, for example, the standard encoder from the AV1 project requires 5721, 5869, and 658 times more computation compared to the x264 (main profile), x264 (high profile), and libvpx-vp9 encoders.
Among the changes in the new SVT-AV1 release:
- Optimization of quality/speed trade-offs has been carried out for presets that set a medium and high quality level. In the "fast-decode 2" mode for presets M3-M10, speed has increased by 15-25% while maintaining the quality level. For presets M0-M2, the quality level has increased by approximately 1% when evaluated using the BD-Rate (Bjontegaard-Delta) method.
- In the "fast-decode 1" mode compared to "fast-decode 0", the number of decoding cycles has been reduced by approximately 10% with only a 1% increase in losses according to the BD-Rate metric.
- On ARM systems, optimizations using Neon and SVE2 extensions have been implemented, allowing a 10-25% increase in the encoding performance of streams with high color depth and resolutions from 480p to 1080p.
- Changes have been made in the SVT-AV1-SPY fork that improve quality perception when using the "tune 0" mode.
- An "avif" mode has been added that consumes fewer resources when encoding static images.
- Unit tests have been improved for code utilizing Arm Neon and SVE2 instructions.
Source: opennet.ru
