@stgraber I have been following this with much excitement. Hopefully I can carve out time this weekend to try it out. One thing I'm curious about that I didn't see covered in the video or blog post (but may have missed) was how it handles things like NVIDIA drivers. I have one deployment of Incus on Ubuntu that uses NVIDIA hardware for GPU acceleration used by a camera server running in an Incus instance. How does the immutable system accept drivers for the additional hardware?

@rossmadness short answer is that it doesn't which is why we're not going into any detail about this stuff.

We do have a kernel config suitable for GPU passthrough to VMs and we can handle GPUs that are supported by the mainline kernel, but NVIDIA is legally tricky to build and distribute.

We certainly couldn't bundle it in the image and having a way to sideload kernel modules after the fact kinda goes counter to the whole immutable thing.

So full GPU passthrough to a VM is your main bet here.

Otherwise for anyone looking to run Frigate without that hassle, I'd recommend an Intel GPU (B50 Pro should even get SR-IOV but a much cheaper A310 will handle encode/decode just fine) and a Coral USB stick for the object tracking models. That's how I have it setup here to handle around 20 cameras.

@stgraber Good to know! Thanks for taking the time to reply.