Can I keep running torch==1.13 with CUDA=11.1?

Hi there!

I wanted to upgrade the torch version I am currently using. I am using a K80 GPU, which, if I understand correctly supports CUDA Driver Version 11.4 and CUDA Runtime Version 11.1. So I am using nvidia/cuda:11.1.1-cudnn8-runtime base image for my docker.

I have been using torch==1.11 before, but recently, I have updated it to torch==1.13. Even though it is not supposed to work with my CUDA 11.1, it seems to be running correctly.

I am wondering why is it the case and if it is possible that something will break if I keep using the newest version of torch.


K80s are still supported in all of our binaries, since we are building for compute capability 3.7-8.6. You should not expect to see any issues.

1 Like