I deployed a VM image from Marketplace with Pytorch and Cuda preinstalled on an Ubuntu OS. At first setup, everything works fine and I'm able to detect NVIDIA GPU from within torch package and my code runs fine.
However, when I restart my VM, the connection between OS and GPU seems to be broken and my VM can no longer detect the graphics card. Only solution so far is to redeploy my VM and start from scratch.
Has anyone faced similar issue or if anyone can help me solving this weird problem?