GPU interaction from container without drivers in host

Hi,

Our application is running in an OS called BalenaOS, that only allows applications to run from containers (without modifying the host OS)… The OS allows to run containers but we cannot install stuff like the “Nvidia drivers”, “nvidia-docker” or “nvidia-container-runtime”… BalenaOS has some implementations for Jetson Devices, but not for the usage of an external GPU.

So, as you can see we are pretty limited in regards to the GPU usage…

We manage to build the drivers independently and load them from the container, but we lost tools like “nvidia-smi” or the creation of the “/dev/nvidia*” entries…

Is there a workaround available for this where we can configure everything required to interact with the GPU from a container alone?

I understand that this question is a pretty exceptional case (and I do not even know if it’s possible), but I know some people that are having same issues within the BalenaOS community…

Thanks in advance!