Hello, we are using a shared V100 GPU server with driver 410 and pytorch 1.7.1 cuda 9.2. And I want to run a project based on torch2.0 with cuda 12. Is it safe to update the nvidia driver to a new version say 535, while keeping my teammates’ projects built on a docker environment with torch 1.7.1 cu92 able to run?
Thanks!
Yes, this should be the case as newer drivers support old CUDA runtimes. However, CUDA 9.2 is quite old that I don’t know what its support is.