GPU Acceleration
Echophrase uses ONNX Runtime’s automatic device selection to pick the fastest available hardware for transcription at runtime. You don’t need to configure anything — the app detects your GPU on first launch and uses it automatically.Supported Hardware
| Platform | Default Accelerator | What you need |
|---|---|---|
| Windows | DirectML | Any DirectX 12 GPU (NVIDIA, AMD, Intel) — no install |
| Windows (Copilot+ PC) | NPU | Snapdragon X or Intel Core Ultra — auto-selected |
| macOS (in testing) | CoreML (Metal) | Apple Silicon (M1+), macOS 13.3+ |
| Linux / WSL | CPU | GPU acceleration on the roadmap |
Why no CUDA?
CUDA support was removed in favor of DirectML. The short version: keeping NVIDIA driver, CUDA Toolkit, and cuDNN versions in sync was a constant source of “works on my machine” support tickets. DirectML ships with Windows, runs on every modern GPU regardless of vendor, and delivers comparable performance for the speech models we run. For the rare workload where CUDA still wins, the right path is the underlying ONNX Runtime — not Echophrase.Verifying GPU Usage
- Open Echophrase Settings
- Look at GPU Acceleration — the top line shows the active device (for example: “Active: Using DirectML (NVIDIA GeForce RTX 3080)”)
- Transcribe a clip and confirm the transcription is fast
Troubleshooting
If GPU acceleration isn’t working:- Update your GPU drivers to the latest from the vendor (NVIDIA / AMD / Intel)
- Confirm your GPU supports DirectX 12 (most cards from 2015 onwards do)
- Check GPU Troubleshooting for platform-specific help