The documentation from nVidia is on installing for a first time. I followed these instructions for the first time install on a fresh Ubuntu 18.04 server. The server has nvidia-docker installed and thus the actual server is free from unneeded packages.
The update to PyTorch 1.7 (from 1.5) requires new drivers. I understand that I can type:
sudo apt update
sudo apt upgradeHOWEVER, since I am 300 miles away from the server, I cannot risk bricking it. My experience with nVidia drivers is: minor tinkering equals days of suffering. Therefore, what is the standard way (or method that's worked many times) to upgrade nVidia drivers?
Specifically, I need to upgrade from 410.48 to 455.32 which I think comes with CUDA 11.1.1
(Please: before down-voting, provide a clear step-by-step link to documentation that shows a safe driver upgrade method that explains why it is safe. I frankly have not found one yet. If I already provided the answer, please pile on the confirmations. Closed CROSSPOST from SO.)
3 Reset to default
