VideoLAN has released dav2d, a fast decoder for the new AV2 codec. This follows their successful dav1d AV1 decoder, which became widely adopted before hardware support caught up. AV2, the successor to AV1, promises up to 25% better compression but requires complex software decoding. dav2d aims to make AV2 practical early, mirroring dav1d’s role for AV1.

The project builds on dav1d’s architecture and tooling, including checkasm for validating optimizations. Current implementations support x86 (AVX2), ARM (NEON), and RISC-V, with features like entropy decoding and motion compensation already functional. While AV2 decoding is five times more complex than AV1, dav2d’s open-source approach allows rapid iteration. The decoder is under a BSD-style license, enabling use in both open and proprietary apps. Development focuses on correctness, platform support, and performance across CPUs.

dav2d’s progress shows the value of open-source tools in advancing codecs. By releasing early, VideoLAN ensures developers can test and optimize AV2 before hardware becomes widespread. This mirrors AV1’s trajectory, where software decoders like dav1d drove adoption years before hardware solutions. The project’s speed reflects lessons from dav1d’s success, combining existing expertise with new AV2-specific work. For AV2 to reach mainstream use, tools like dav2d are critical—proving that open implementations can bridge the gap between specification and real-world deployment.