Skip to main content

Hardware Acceleration & Raspberry Pi 5

Kinema supports hardware-accelerated video decoding to reduce CPU load and power consumption on small, low-power single-board computers (SBCs) and servers.

Raspberry Pi 5 V4L2 HEVC DRM Acceleration​

The Raspberry Pi 5 (Broadcom BCM2712 SoC) features a dedicated hardware stateless HEVC decoder exposed through Video4Linux2 (v4l2-request / DRM), but lacks a hardware video encoder:

Automatic Detection Mechanics​

In crates/kinema-core/src/platform/unix.rs:

pub fn has_v4l2_hevc_decoder() -> bool {
    if let Ok(entries) = std::fs::read_dir("/sys/class/video4linux") {
        for entry in entries.flatten() {
            if let Ok(name) = std::fs::read_to_string(entry.path().join("name")) {
                if name.trim().ends_with("hevc-dec") {
                    return true;
                }
            }
        }
    }
    false
}

When has_v4l2_hevc_decoder() returns true:

  1. Kinema's binary resolver probes the system's distribution ffmpeg (/usr/bin/ffmpeg) for DRM device capabilities (decoder=hevc with device drm).
  2. While standard tasks use Kinema's bundled static ffmpeg binary, HEVC decoding tasks automatically switch to bins.hevc_drm with flags: 
    /usr/bin/ffmpeg -hwaccel drm -hwaccel_device /dev/dri/renderD128 -c:v hevc_v4l2request -i input.mkv ...
  3. This eliminates ~35% of the total CPU load on Raspberry Pi 5 boards, preventing thermal throttling during playback.

Intel QuickSync & VAAPI (Linux x86_64)​

On Intel and AMD systems with /dev/dri/renderD128 available:

  • Kinema can pass -hwaccel vaapi -hwaccel_output_format vaapi when VAAPI drivers are detected.
  • In Docker environments, ensure the render device is passed: 
    devices:
      - /dev/dri:/dev/dri

NVIDIA NVENC / NVDEC​

On Linux and Windows systems equipped with NVIDIA graphics cards:

  • Kinema can invoke -hwaccel cuda for hardware-accelerated video decoding.
  • Set prefer_hardware_accel = true in kinema.toml.