Would a massive cache make much difference?

[VictordeHolland]

Originally Posted by NookieN

I have the chip in question and it is fast but I didn't find any compelling reason to run more than one worker. With all 22 cores active they will run at 2.6GHz. For 4M the peak throughput was about 725i/s (1.38ms) on all 22 cores. 2 workers is noticeably slower and with 3 workers and up the total throughput was always around 300i/s. I can rerun and post benchmarks or gwnum if anyone is really curious.

So you're saying with 1 worker using 22 cores you get 725 iter/sec and with 3 workers (each using 7 cores?) you get a total throughput of 300 iter/sec, so about 100 iter/sec for each worker? That is a huge difference! Much bigger difference than I expected!

[kruoli]

Originally Posted by NookieN

I can rerun and post benchmarks or gwnum if anyone is really curious.

Yes, please!

[NookieN]

Originally Posted by VictordeHolland

So you're saying with 1 worker using 22 cores you get 725 iter/sec and with 3 workers (each using 7 cores?) you get a total throughput of 300 iter/sec, so about 100 iter/sec for each worker? That is a huge difference! Much bigger difference than I expected!

Yes, at least that's how I interpret it. The results don't state the number of cores used per worker, so I would assume it's using 11 cores for 2 workers, 7 for 3, and 5 for 4.

Results for 4M attached. I find it amusing that the cache layout is longer than the bench results.

[Prime95]

Thanks! An excellent demonstration of how memory-bound prime95 is.