Skip to content

bench: decode NVENC's HEVC and AV1 streams on the Mac and the iPhone #82

Open

Description

@with-developer

Why

The HEVC comparison (#80) decodes streams that the Mac and the iPhone encode themselves with VideoToolbox. A Moonlight client decodes its host's streams instead: NVENC's HEVC, and AV1, which Apple's recent devices decode in hardware but cannot encode. No AV1 encoder is installed on the Mac, so AV1 has not been timed at all.

What

  • Record short streams on a PC with NVENC (an RTX 4070 Super is at hand): HEVC and AV1, low-latency settings as Sunshine uses them, at Moonlight's default bit rates for 1080p, 1440p and 4K at 60 frames per second, of the benchmark's pictures or of a game.
  • Decode them in the benchmark app and its Mac front-end through VideoToolbox, in the same way as the device's own HEVC streams, beside emberwave.
  • Note what changes against the device's own streams.

Activity

  1. added a commit that references this issue on Nov 18, 2026
  2. with-developer commented on Nov 18, 2026

    @with-developer
    OwnerAuthor

    First stage is in main (#167, BENCHMARKS.md "Against AV1 and HEVC in Chrome (stand-in for NVENC)", D67 follow-up). The PC and its NVENC were not at hand, so Chrome 157's WebCodecs encoders on the M3 Max stood in: HEVC and H.264 in hardware (VideoToolbox), AV1 in software, decoded by VideoDecoder, on two synthetic sequences (the benchmark scene, and a copy with fresh noise in every frame) at three sizes and two bit rates each. emberwave is measured at the bytes each stream used.

    • At the same bytes the codecs with predicted frames lead in PSNR: 15.4 to 17.7 dB of luma on the scene, 3.1 to 4.6 dB on the noisy copy. emberwave is given more bytes than the predicted frames cost (a key frame is 12 to 89 % of a codec stream), so the comparison favours emberwave. The scene's own grain caps the codecs near 46 dB.
    • VideoDecoder took 0.9 to 5.3 ms a frame at Moonlight's rates on the scene (hardware decoders) and up to 12.6 ms at emberwave's bytes on the noisy copy.
    • Chrome's encoders did not fill the asked rate on the scene (AV1 3.8 of 20 Mbit/s at 1080p); six rows did not run (hardware H.264 returned no frames at 336 and 576 Mbit/s, a hardware HEVC stream at 1440p could not be decoded).
    • For D67's chroma weights the numbers say little: emberwave's chroma PSNR sits further above its luma PSNR than the codecs' in most rows, which is the direction of the tilt, but PSNR counts every band alike and the pictures are synthetic. The weights stay.

    Still open here: NVENC's own streams (the PC), moving game footage, and the iPhone.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    perfPerformance

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions