How to use global CUDA installation with PyTorch

I have installed CUDA Toolkit 10 and cuDNN 7.4.x globally for use with my TensorFlow environment but when it comes to PyTorch so my question is can PyTorch GPU version be installed in a conda environment and the environment utilities the global CUDA 10 installation (from NVIDIA’s website) as opposed to the conda provided CUDA package?

Yes it can be, but you’d have to compile from source to link against these versions.

But since I am beginner, could you provide some guidance on how to do that on Windows 10?

Instructions for that, incl. Windows, can be found here: