Thank you very much.
I would like to install Pytorch on windows.
Which Pytorch version is suitable for me?
CUDA version is 9.0 on my laptop computer.
The latest version of Pytorch is v.1.6.0, which only supports CUDA versions 9.2 or newer versions.
The version of Pytorch v.1.1.0 can support CUDA version 9.0.
So I am not sure whether I should install the version v1.1.0 of Pytorch with CUDA or the version v1.6.0 Pytorch No CUDA?
Which version is suitable for me?
If I install Pytorch v.1.1.0, then if the codes are based Pytorch v.1.6.0 in one article, will I meet troubles when I try to copy the codes and do numerical experiments?
The latest PyTorch release is currently 1.7.1 and I would recommend to stick to it instead of falling back to old versions.
Your local CUDA installation will not be used to run PyTorch, as the conda binaries and pip wheels ship with their own CUDA runtime. You would have to make sure to use the appropriate NVIDIA driver on your system to use the CUDA runtime.
Your local CUDA toolkit would be used, in case you want to build PyTorch from source or any custom CUDA extensions.
Thank you very much.
In this web page, there are two versions, CUDA Toolkit 11.0 and CUDA Toolkit 11.0 Update1 . CUDA Toolkit Archive | NVIDIA Developer
I am confused. If I want to install Pytorch version 1.7.1 with CUDA, Which CUDA should I choose?
If you want to build custom CUDA extensions, I would recommend to stick to the matching one which ships with the binaries, so 11.0.