I ran an nvidia docker container on my Jetson Nano following this hello world example: https://github.com/NVIDIA/nvidia-docker/wiki/NVIDIA-Container-Runtime-on-Jetson#hello-world
sudo docker run --runtime nvidia --network host -it -e DISPLAY=$DISPLAY -v /tmp/.X11-unix/:/tmp/.X11-unix nvcr.io/nvidia/l4t-base:r32.2
javascript:void(0);
I was surprised to see that inside the container, my /usr/local/cuda directory was the same /usr/local/cuda directory as my host. How did that happen? I didn’t explicitly map that directory into the container.
It seems to be caused by specifying --runtime nvidia. What other surprises await me when I do --runtime nvidia (in terms of directories from the host that are mapped through to the container)? What other directories are mounted? Is this documented somewhere?
I also did a docker inspect to see binds or volumes, but the only thing I noticed is the binding of the /tmp/.X11-unix folder for the X socket. (Perhaps I’m overlooking something.)