Skip to main content

Hardware Acceleration

Alchemist detects hardware automatically at startup, actively probes every plausible backend/codec candidate, and selects a single active device/backend with a deterministic scoring policy. Override in Settings → Hardware.

Detection flow (auto mode)

  1. Discover plausible candidates
    • Apple VideoToolbox on macOS
    • NVIDIA NVENC when NVIDIA is present
    • Intel / AMD render nodes on Linux via /sys/class/drm/renderD*
    • AMD AMF on Windows
  2. Actively probe every candidate encoder with a short FFmpeg test encode
  3. Group successful probes by device path / vendor
  4. Choose one active device/backend using codec coverage, backend preference, and stable vendor ordering
  5. Fall back to CPU only if no GPU probe succeeds and CPU fallback is enabled

Encoder support by vendor

VendorAV1HEVCH.264Notes
NVIDIA NVENCRTX 30/40Maxwell+AllBest for speed
Intel QSV12th gen+6th gen+AllBest for power efficiency
AMD VAAPI/AMFRDNA 2+ on compatible driver/FFmpeg stacksPolaris+AllLinux VAAPI / Windows AMF; HEVC/H.264 are the validated AMD paths for 0.3.0
Apple VideoToolboxM3+M1+/T2AllBinary install recommended
CPUAllAllAllAlways available

Hardware probe

Alchemist probes each encoder at startup with a test encode using a standardized 256x256 lavfi input.

See results in Settings → Hardware → Probe Log. The UI shows:

  • the selected device/backend reason
  • probe counts (attempted / succeeded / failed)
  • per-probe summaries for success and failure
  • full FFmpeg stderr for failed probes

On Linux, explicit device paths only apply to render-node backends such as VAAPI and QSV.

Vendor-specific guides