What's the relationship between GPUs and package management?
Most deep learning these days requires CUDA acceleration to enable GPU / TPU for the libraries (i.e. PyTorch, Tensorflow), which is an absolute nightmare to set-up.
NVIDIA has a bunch of docker containers which make this slightly less painful. Or maybe it’s just a different kind of pain.
These definitely help.. but still painful.
I wanted to try SD on my machine, but I just couldn't get CUDA to work. Mind, the the openvino based CPU implementation works just fine (pip install -r requirements.txt was sufficient) and given my CPU and GPU (a 10 core 10850k and and old 1070), there's isn't probably much to gain to switch to the GPU other than power usage.
I guess the old "NVidia -vs- Linus" battle on kernel modules is still being fought....