FFmpeg 9.0 Release Notes

FFmpeg 9.0 "Lei" is a major release that expands hardware acceleration capabilities, adds support for new multimedia formats, and refines the project's deep learning integration. This release arrives approximately four months after FFmpeg 8.1.

Hardware Acceleration and GPU Enhancements

FFmpeg 9.0 significantly expands its hardware-accelerated processing, particularly for AMD and Vulkan-based workflows.

AMD Media Framework (AMF) Updates

  • HDR Capabilities: The AMF Color Converter (vf_vpp_amf) now features extended HDR capabilities.
  • Frame Rate Conversion: A new AMF Frame Rate Converter filter (vf_frc_amf) has been added.
  • Memory Mapping: The release introduces support for AMF hardware memory mapping.

Vulkan and CUDA Integration

  • Vulkan Filters: The v360_vulkan filter is now available for spherical video processing.
  • CUDA Filters: A new transpose_cuda filter has been added for hardware-accelerated image rotation.
  • Vulkan Hardware Acceleration: New hardware acceleration (hwaccel) support has been added for APV via Vulkan.

Other Hardware Accel

  • VideoToolbox: ProRes RAW now has VideoToolbox hardware acceleration support.

New Codecs, Muxers, and Demuxers

This version adds support for several niche and modern formats to ensure broad compatibility across diverse hardware and software ecosystems.

  • Animated WebP: FFmpeg 9.0 adds both an Animated WebP decoder and demuxer.
  • Playdate: Support for the Playdate video encoder and muxer has been introduced.
  • LCEVC: The MP4 muxer now supports LCEVC track muxing.
  • HE-AAC 960: Decoding support for HE-AAC 960 (DAB+) is now available.
  • Dolby Vision: A new bitstream filter has been added to split Dolby Vision multi-layer HEVC.
  • SMPTE 2094-50: The release adds support and passthrough for SMPTE 2094-50 metadata.

Deep Learning and Backend Changes

FFmpeg continues to integrate AI-driven processing through its DNN (Deep Neural Network) framework.

  • ONNX Runtime: The ONNX Runtime DNN backend now supports the GPU execution provider, allowing for faster inference of neural network filters.

Deprecations and Removals

To maintain codebase health, FFmpeg 9.0 removes several legacy components:

  • CELT: Support for CELT decoding and ogg/celt parsing has been removed (this does not affect Opus CELT).
  • NVENC: Deprecated NVENC options and support for SDK versions prior to 11.1 have been removed.

Community Insights and Technical Discussion

Users and developers on Hacker News highlighted the enduring importance of FFmpeg as a foundational piece of open-source infrastructure. While the release is praised, some users noted the steep learning curve of the filter system.

"ffmpeg also has many really powerful filters, but these are very confusing to use IMO and not elegant at all. I'd wish we could use some kind of simple meta-language... to think of multimedia data as tangible to manipulation as if it were an object oriented system or datastream system."

Other discussions touched upon the project's commitment to extreme efficiency, with users referencing interviews with the project's lead engineers regarding the use of hand-rolled assembly code to maximize performance.

Sources