Dear Nizamov, dear Ben, dear Meep-users! We are working at Intel quad core xeon processors and I can confirm only little acceleration when using more than 1 core on the same CPU (20 - 30%).
To my knowledge the limiting factor is not the computational power of the processors, it's the interface to the system memory. As in Intel processors all cores share the same front side bus, execution times decrease very little when starting more than one process on the same CPU (not core). This should be better on AMD CPUs and the new Intel Core i7 series. Ben, are you working on AMD CPUs? Is there already any experience about the performance with the new Intel Core i7 CPUs? Best regards, Roman -----Ursprüngliche Nachricht----- Von: meep-discuss-boun...@ab-initio.mit.edu [mailto:meep-discuss-boun...@ab-initio.mit.edu] Im Auftrag von Benjamin M. Schwartz Gesendet: Montag, 12. Jänner 2009 18:22 An: Nizamov Shawkat Cc: meep-discuss@ab-initio.mit.edu Betreff: Re: [Meep-discuss] MPI efficiency Nizamov Shawkat wrote: > So the first question is - what mpi efficiency do you observe on your > systems? I have observed excellent acceleration. Dual-core systems have been near-linear, and running on 8 machines with a GigE interconnect I got about 6x acceleration. The important thing is to make sure that your problem size is large enough. The website suggests that MPI is only useful for 3D problems, and I would add that it is best for problems that are at least 100 MB in size. If your problem size is already very large, then it's possible that there is a misconfiguration in your system. --Ben _______________________________________________ meep-discuss mailing list meep-discuss@ab-initio.mit.edu http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
