That will hurt latency and latency is very important for good tensor-parallelism performance
I think I need one now that I have four - reduce is ring-oriented and still works I believe, but IIRC you only get 200gbps if you use _one_ of two connectx ports.
alt Hacker News
I think I need one now that I have four - reduce is ring-oriented and still works I believe, but IIRC you only get 200gbps if you use _one_ of two connectx ports.