graphcore 800MB in-process memory vs groq 230MB
graphcore 47.5T B/s in-process memory bandwidth vs groq 80TB/s
grachore 320GB/s chip to chip networking bandwidth vs groq 480GB/s

从memory hierarchy 到chip 2 chip 这条通路上，graphcore和groq的设计思路是一致的，如果groq能成功，graphcore也能。