Hacker News Reader

Outside of Macs, llama.cpp running fully on the cpu is more than 10x slower than a GPU.

But having 32 real cores in a cpu is so much cheaper than having mumtiple gpus. RAM is also much cheaper as VRAM.