Hacker News Reader

Stable Diffusion runs pretty fast on Apple Silicon. Not sure if that uses the GPU though.

I think one reason in this particular case may be the 4-bit quantization.

Quantization is the answer here. CPU running the large models at 16 bits (which is actually 32, because CPUs mostly do not support FP16) would be really slow.

Reply on news.ycombinator.com