Torch.cuda.is_available() returns False with cuda10.0

whisht · April 26, 2019, 11:57am

torch.cuda.is_available() returns False with cuda10.0, but True with cuda8.0. So I am confused with this problem.

I have installed cuda with different version(8.0 as well as 10.0) and the corresponding cudnn, but the returned value is opposite. Is this relative to the NVIDIA gpu driver version(which is 375.26)? Does this makes the cuda10.0 not working?

kshitij · April 26, 2019, 12:09pm

I strongly believe that you are correct. The issue has to do with the CUDA and driver incompatibility.

cuda_compatibility

Reference : CUDA Compatibility :: NVIDIA Data Center GPU Driver Documentation

So you may have to upgrade drivers for CUDA 10 support

joellerena · October 5, 2019, 1:28pm

Also with torch.cuda.is_available () had false.
But when installing the Nvidia driver to the most updated version 436.48, True is displayed. I previously updated Pytorch to 1.2.0.
I have windows 10 and anaconda.