I wonder if someone has run into this before. I compiled one of our CFD
solvers on OS X using MPI, and have been testing it on a variety of G4 and
G5 systems. When I run across multiple machines using only one processor
per machine, the performance is great. When I use dual processors, however,
the problem takes forever to setup (initial communication overhead of the
flow solver I assume) and then cranks OK once the iterations start.
I was expecting some penalty when running dual processors as they compete
for bandwidth, but it seems way too slow. In fact, I could manually run
about a dozen single CPU cases in the time it takes for one dual-processor
case to get going.
Any ideas or suggestions??
Thanks,
Craig
--
Dr. Craig Hunter
NASA Langley Research Center
AAAC/Configuration Aerodynamics Branch
email@hidden (new!!)
(757) 864-3020
(Dual G4 - OS X)
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Scitech mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden
