You need latency for token parallelism, not bandwidth. Hence actual RDMA that by... | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

Lwerewolf 32 days ago | parent | context | favorite | on: New Mac Studio with M5 Max and M5 Ultra

You need latency for token parallelism, not bandwidth. Hence actual RDMA that bypasses the software TCP stack (ROCe or whatever).

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact