3 ms·
If you are memory throughput bound, and both processors are working independently (think embarrassingly parallelizable problems) you'd probably benefit from the
by Terribledactyl 8y ago
If you are memory throughput bound, and both processors are working independently (think embarrassingly parallelizable problems) you'd probably benefit from the total throughput change.
Otherwise it's pretty easy to get worse performance due to numa issues
- zrm 8y agoEPYC is already four NUMA nodes per socket though.