Cannot start with nvidia.runtime=true I’m on debian 12 “Bookworm”, I have install proprietary nvidia-driver that is at Driver: nvidia (535.247.01) as reported by incus info –resources I’m not sure if I’m missing a library here, but even nvidia-container-cli info is failing with nvidia-container-cli: initialization error: nvml error: driver not loaded I have nvidia-container-toolkit 1.17.6. Do I need CUDA or something? It seems very superfluous as I’m just using it for transcoding 
You need to have the Nvidia driver on the host, no need for the client libraries on the host but be my guest to figure out how to avoid installing those packages.
If everything is fine on the host, the command nvidia-smi (on the host) should show that the GPU is running with the binary drivers from NVidia.
Then, when you launch a container with nvidia.runtime=true, then Incus should provide the proper container image to the container. In the container run nvidia-smi and you should be getting a matching output.
You can configure your container to use fewer capabilities from the GPU. The default is compute,utility. See https://linuxcontainers.org/incus/docs/main/reference/instance_options/#instance-options-nvidia for more. If unsure, start with all.
Cannot verify the nvidia-container-clioutput at the moment.
I solved the issue, I did not install nvidia-smi as it wasn’t listed as a requirement but after it was installed everything seems to be fixed. I wonder why nvidia-smi isn’t listed as a requirement if it indeed is one.