2 ms·
I wonder if it would be better to scale down cluster paradigms (MPI stuff), rather than trying to somehow scale up shared-memory paradigms.
by bee_rider 4y ago
I wonder if it would be better to scale down cluster paradigms (MPI stuff), rather than trying to somehow scale up shared-memory paradigms.
- JonChesterfield 4y agoDropping cache coherency is a big lever for performance. That's definitely more annoying to program against than a magically coherent model though.
- bee_rider 4y ago“Big lever” is a nice expression, successfully expresses both the power and unwieldiness of a thing.