> I have been testing speed with different number of cores. And to my > surprise I must tell you, that when more than 5 cores are used - the > speed drastically drops down! > > For instance with 4 cores is is steady 200it/sec. > On 6-8 cores it is 1 it/sec, and occasionally (less than 5% of the > time) the speed jumps up to 2000it/sec). It is visible also in > OpenGL window - everything is stalled, and after 10 seconds it > suddenly moves extremely fast, for 0.5 of sec. Then again everything > stalls.
That is quite possible. I also had best results with 5 cores on Opterons, but Xeons were getting faster up to 8 cores (about 5.5x). Depends on memory bus and other stuff I don't understand that much. The jumping speed could be also explained bu the GUI thread blocked by computation, so it updates only very rarely and shown nonsense values. Note that there are still some global locks (such as when creating interaction) and non-parallel parts like the collider. See https://www.yade-dem.org/wiki/Triaxial_Test_Parallel and https://www.yade-dem.org/wiki/Performance_Tuning, in most cases 3-4 cores give the best performance. Cheers, v. _______________________________________________ Mailing list: https://launchpad.net/~yade-dev Post to : [email protected] Unsubscribe : https://launchpad.net/~yade-dev More help : https://help.launchpad.net/ListHelp
