Maybe this will help. I don’t have an Nvidia card to test that, though.
When you set nvidia.driver.capabilities: all
and nvidia.runtime: "true"
in your config, the /proc/driver/nvidia/gpus/0000:01:00.0
from the host will be mounted to /dev/nvidia0
folder in the container. Simple symlink in the container should work (there’s a systemd service in the link above that automates this):
sudo mkdir -p /proc/driver/nvidia/gpus && ln -s /dev/nvidia0 /proc/driver/nvidia/gpus/0000:01:00.0
The other solution is to skip nvidia.driver.capabilities: all
and nvidia.runtime: "true"
and install all the Nvidia packages directly in the Incus container.