News for hackers. Stuff that matters.
Re:Intel Doubles Down on Doubled Up Xeons for HPC by rbanffy on Monday November 12, 2018 @10:32AM (#18433165)
It depends on your problem. If your workload spends a lot of time waiting for a cache miss, halving the number of threads and spreading them across cores will increase available cache for each thread and increase your throughput substantially. If the working data is larger than the cache, additional memory channels will make your life much better, so getting the CPU with the most memory interfaces will be more important than getting the one with more cores, threads, SIMD pipelines or GHz. I would say that serving web requests from one of these or simulating quantum computers is not the best cost/benefit you'll find, but if you need to do something like CFD, it seems like a good match.

I may be off a bit, but the last two generations of Xeon Phis could emulate a general 10-12 qubit quantum computer without leaving the HBM memory on the processor module and hitting the external memory buses. I'm curious if anyone has tried that.

There are no comments on this comment yet.