I’m interested in gpu-gpu memory access performance presented in the following presentation :
It states P2P communication is fully supported on a 4x GPU, 1 IOH Setup, and it looks like GPU0 can talk to GPU1 through the switch.
This is really great news, but does the switch need to be an NF200? I’d love to build the setup described in this slide, with an LGA2011 processor, but I can’t find a compatible motherboard yet!
Is any way to build this setup now or should I wait?