Anyone else find that using:
A) Windows
and
B) Python
to do GPU based machine learning is an absolute mess.
I previously had Ubuntu in a dual OS boot setup, and you could do a 1 line install that took care of everything: https://lambdalabs.com/lambda-stack-deep-learning-software
I have now moved everything to Windows, and I am pulling my hair out (what’s left of it anyway) trying to get some old notebooks up and running. The myriads of CUDA, CuDNN, drivers, python, keras and TF version and incompatibilities is a real nightmare.
I bought a 2080TI for a reason, and trying to get it to scream doing CNN is currently not happening.
I can see why cloud based distros, with everything pre-built are so useful.
Makes we want to go to that dual boot again…