With third gen Threadripper, we’ll see better IPC and a few improvements supporting the latest technology, but half the cores will still be limited in the way they are forced to indirectly access RAM.
Direct core access to RAM will certainly speed up future generations of AMD CPU’s, but what other elements are needed - or - what shortcomings need to be addressed, in order to maximize performance on future 64+ core count CPUs?
The amazing part there is that it is apparently only about 20% behind on Ryzen for 256bit. Considering it is lower clocks, IPC, better performance per watt and
while having to add 128bit outputs… Damn, Intel are holding out but holding out like they are slipping off a cliff and their fingernails are the last thing holding them.
8 Channel memory I think latency is a problem that will be halfed on the next node when they are able to have the space. But we could see optimization on scheduling processes for Threadripper.