[Discussion] How to use CUDA SDK installed on host?

[chrismile]

Hi,
I run Fedora 39 on my base system, which has GCC 13.2 that is incompatible with all current versions of nvcc. Thus, I am trying to use distrobox to install a Ubuntu 22.04 container.

When I create the container with distrobox create --nvidia --name ubuntu-nv --image ubuntu:22.04, the directory /usr/local/cuda-12.1 from the host system seems to be made available to the container, but for some reason it does not contain the sub-folder bin present on the host system.

When I create the container with distrobox create --nvidia --name ubuntu-nv --image ubuntu:22.04 --volume /usr/local/cuda-12.1:/usr/local/cuda, no CUDA folder is made available in the container for some reason.

Am I doing something wrong here? I am aware of the possibility to just access all files of the host via /run/host.