[DGD] Hydra

[Felix A. Croes]

Dread Quixadhal <quixadhal at gmail.com> wrote:

>[...]
> http://quixadhal.ddns.net/~quixadhal/hydra/
>
> Interestingly I never saw a CPU utilization of more than 50%, even when telling it to do 32 threads,
> and a single CPU core never got above 70%.

Thanks for running the tests!  The results are interesting because the
CPU is quite recent.

The reason 8, 16 and 32 task runners produce pretty much the same results
is that there are only 8 hardware threads available on your box, so
Windows has to share hardware threads between the task runners.  4 task
runners is optimal because hyperthreading doesn't do much for Hydra.
When running normal code instead of a synthetic benchmark, you would
probably see an even bigger slowdown with 8+ task runners.

You never see high CPU utilisation because the test was written to
make lock contention the main constraint on concurrency.

Regards,
Felix Croes