Distributeddataparallel across two machine with different version of Pytorch and CUDA

Hello, I am trying to set up distributed learning across two laptops using distributeddataparallel. One computer has CUDA 10 and the other one has CUDA 11. Will this create problem during training?

I haven’t tried mixing different CUDA and NCCL versions in DDP runs, but guess it might yield unexpected behavior or other issues. Especially, if you are also mixing different PyTorch releases.