> because they feature 800Gbps of memory bandwidth. About twice what's available on the top end of M5

Ouch, about half of the memory bandwidth of a dedicated GPU though :/ Running LLMs on Apple hardware still doesn't make any sense to me.