Hallo,

I'm using multi-threading on different machines with 1, 2 or 8 CPUs 
(deal-version 5.2.0).

If I stop the calculating time for matrix-vector-multiplication with 'vmult' on 
the different machines with
'clock_t' and 'time_t' , I'll observe that 8 CPUs need a  few times as long as 
2 CPUs.
I have tested it with 196698 dofs.

Assembling system-matrix and right-hand-side works fine (stopped with difftime 
(end, start), all on the same machine):
1 CPU  : 63 sec  // n_threads = 1;
2 CPUs: 31 sec  // n_threads = 2;
4 CPUs: 15 sec  // n_threads = 4;
8 CPUs:  9 sec  // n_threads = multithread_info.n_default_threads;

Is it possible to set the number of threads for 'vmult' too?
Or could there be another mistake in my code?

Thanks for help,
Jörg
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii

Reply via email to