GPU Acceleration

GPU support in Facelock is runtime-only -- no special build flags or recompilation needed. Install a GPU-enabled ONNX Runtime package for your hardware and set execution_provider in the configuration.

Setup

1. Install a GPU-enabled ONNX Runtime

GPU Vendor	Arch Linux Package	Other Distros
NVIDIA	`onnxruntime-opt-cuda`	Install CUDA toolkit + ONNX Runtime with CUDA provider
AMD	`onnxruntime-opt-rocm`	Install ROCm runtime + ONNX Runtime with ROCm provider
Intel	`onnxruntime-opt-openvino`	Install OpenVINO runtime + ONNX Runtime with OpenVINO provider

On Arch Linux:

sudo pacman -S onnxruntime-opt-cuda      # NVIDIA
sudo pacman -S onnxruntime-opt-rocm      # AMD
sudo pacman -S onnxruntime-opt-openvino  # Intel

2. Set the execution provider

Edit /etc/facelock/config.toml:

[recognition]
execution_provider = "cuda"    # or "rocm" or "openvino"

3. Restart the daemon

facelock restart

4. Verify

facelock bench warm-auth

Compare latency with execution_provider = "cpu" to confirm GPU acceleration is active.

How it works

Facelock uses the ort crate with the load-dynamic feature. At startup, it loads libonnxruntime.so from the system library path. If a GPU-enabled ONNX Runtime is installed, it provides CUDA/ROCm/OpenVINO execution providers automatically. The execution_provider config selects which provider to register.

If the requested provider is not available (e.g., CUDA requested but only CPU ORT installed), Facelock falls back to CPU with a warning.

Supported providers

Provider	Config value	Status
CPU	`"cpu"`	Default, tested
CUDA (NVIDIA)	`"cuda"`	Config ready, requires GPU-enabled ORT
ROCm (AMD)	`"rocm"`	Config ready, requires GPU-enabled ORT
OpenVINO (Intel)	`"openvino"`	Config ready, requires GPU-enabled ORT

systemd note

The systemd service has MemoryDenyWriteExecute=yes commented out because GPU inference runtimes (CUDA, TensorRT) use JIT compilation which requires writable+executable memory pages. If you are using CPU-only, you can re-enable this directive for additional hardening.

Troubleshooting

"Failed to load execution provider": The GPU-enabled ONNX Runtime package is not installed or libonnxruntime.so does not include the requested provider.
Slower than CPU: Ensure the GPU driver is loaded (nvidia-smi for NVIDIA, rocm-smi for AMD). Small models like SCRFD 2.5G may not benefit from GPU due to transfer overhead.
Daemon crashes on startup: Check journalctl -u facelock-daemon for ORT initialization errors. GPU memory allocation failures are the most common cause.

Facelock Documentation