Build pytorch-gpu for different gpu archs

I build torch-1.6.0a0+b31f58d-cp36-cp36m-linux_x86_64.whl on my Tesla V100 machine, when I install the wheel package on my Tesla T4 machine, I met the warning like this:

Tesla T4 with CUDA capability sm_75 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_70.

I wonder how should I build pytorch-gpu for different gpu archs?

Github issue here.

I’ve answered in the GitHub issue.

1 Like