After spending quite some time experimenting with different setups, I’m thrilled to share my successful experience running Frigate NVR on a Windows 11 mini PC. My journey started with the Raspberry Pi 5, but the inference times were just too high for my liking. I decided to give Windows a shot, and boy, am I glad I did!
I set up Frigate using WSL 2 and Docker Desktop. The process was smoother than I anticipated, though there were a couple of hurdles along the way. One thing I noticed was the need to manually correct the model path in the config.yaml file. Initially, it pointed to an incorrect 1KB file, but after some digging, I found the correct 160MB model in a subfolder named 8.5.3. Adjusting the path there made all the difference!
Here’s a quick rundown of my setup:
- Hardware: Intel i9 10th gen mini PC with an external RTX 3060 12GB GPU.
- Software: WSL 2, Docker Desktop, and the latest stable Frigate container with TensorRT support.
I followed these steps:
- Created essential folders:
~/frigate/config,~/frigate/media, and~/frigate/models. - Configured
config.yamlwith my camera inputs and detection settings. - Ran the Docker container with the necessary GPU and volume mappings.
One thing I’d emphasize is ensuring Docker has access to your GPU. The --gpus all flag and setting NVIDIA_VISIBLE_DEVICES=all were crucial for performance. Also, don’t forget to adjust the model path if Frigate doesn’t automatically detect it.
Now, everything runs smoothly! The inference times are significantly better than on the Raspberry Pi, and the integration with Home Assistant is seamless. I can’t recommend this setup enough for anyone looking to leverage their Windows hardware for home security.
If you’re considering this setup, feel free to reach out with any questions. Happy automating! 