Agreed. Sorry if I gave the impression of being pro-nVidia. I am not. But the reality is that when Tensorflow and Pytorch came to be there was no alternative.

Agreed. Sorry if I gave the impression of being pro-nVidia. I am not. But the reality is that when Tensorflow and Pytorch came to be there was no alternative. Now you need to jump through hoops to make it work with non CUDA hardware. Additionally, while drivers play a role, I think the main difference is in the computing libraries (CUDA vs ROCm)