>>85979745Basically, Intel is
>lolAMD are giga super mega uber retards who can not design a functional framework for their GPUs,
which leaves Nvidia as the only ones producing GPUs with usable API for Machine Learning (CUDA).
Nvidia in their Biblical Jewishness refuse to make a card with enough VRAM for AI bullshit, so that leave two options
>Buy a dedicated AI GPU with ~80 gigs of VRAM for 20k (if you can find one)or
>Buy four 4090s for ~8-10k and make a server clusterFuck GPU manufacturers
