This might be a super dumb question, but are super computers worth it? Meaning anything that requires custom hardware instead of just groups of coordinated commodity hardware. It seems you can get maaaybe 5x (10x?) current performance but at a greater (100x?) cost multiple. Or is this extra spend on custom supercomputer hardware what effectively sponsors the research that allows Moore's law to continue?
it looks like 'hpc ethernet' is really just a classic cray memory network design with hop-by-hop retransmission and rigid flow control.<p>so really a port can fall back to supporting ethernet? maybe it would have been easier to just put an adapter in there?<p>just wondering what the real meat is. i do think convergence with non-supercomputer systems is a great idea and should help quite a bit with NRE