Dav2d

Original Article ↗ Hacker News Discussion ↗

A new open-source software decoder for the next‑generation AV2 video codec is drawing interest for promising about 25% better compression than AV1 at the cost of roughly five times higher decoding complexity. Commenters weigh the practical trade-offs: increased CPU demands, lagging hardware support, and security risks from performance‑critical C/assembly implementations versus safer Rust-based ports that are still somewhat slower. There is also debate over long‑term patent uncertainty around “royalty‑free” codecs and whether incremental gains in compression justify the encoding/decoding burden and potential device obsolescence.

Patent and Licensing Concerns

  • Multiple comments ask whether AV2 can avoid the kind of patent-pool issues seen with AV1 (e.g., Sisvel, Dolby/Snap case).
  • Some argue that lawsuits and license offers do not prove patents are valid; only courts can.
  • Others point out that Sisvel has not yet enforced AV1 patents in court and previous VP9/AV1-related patents have been invalidated.
  • Consensus: no codec can fully avoid patent-pool claims; legal risk is inherent.

Compression vs Complexity

  • AV2 is described as ~25% more efficient than AV1 (e.g., same quality at ~75% of the bitrate).
  • Decoding is said to be ~5× more complex, interpreted as ~5× more computationally expensive.
  • Early encoders are extremely slow; commenters expect performance to improve with time, but still be slower than AV1/H.265.

Hardware, Practical Use, and “Obsoleting” Devices

  • AV1 is already widely deployed (e.g., major streaming platforms), with both hardware and fast software decoders (dav1d).
  • Concerns that AV2 will be too heavy for many devices, especially older/cheap ones and TVs; interaction lag is already noticeable with AV1 on some hardware.
  • Others note streaming services already maintain many codec/bitrate variants; AV1 will remain in use for years and new codecs don’t immediately obsolete old devices.
  • For very high‑volume content, the 25% gain is considered economically important.

Language Choice: C/ASM vs Rust

  • Strong argument that performance is paramount for video codecs; heavy use of hand-written SIMD assembly is standard.
  • C is preferred as a thin layer around assembly; Rust is seen as safer but harder to push to absolute performance, especially with bounds checks.
  • Counterarguments say safety should be prioritized, given untrusted input, and that Rust ports of AV1 (and now AV2) show modest overhead (claims range from ~5% to ~20%).
  • Debate over whether sandboxing unsafe C/ASM is “cheap enough” versus using memory-safe languages.

Miscellaneous Topics

  • Curiosity about AV2’s behavior at very low bitrates, with nostalgic references to tiny “movie in a few MB” encodes.
  • Discussion of naming (AV1/AV2, dav1d/dav2d) and jokes about future versions.
  • The original blog post suffered a “HN hug of death,” requiring use of web archives.
  • Some speculate that rapidly rising decode complexity suggests approaching the limits of hand-engineered codecs, with neural-codec approaches as a possible future direction.