[Beowulf] Re: Opteron 275 performance

[Steve Cousins]

On Wed, 27 Jul 2005, Ben Mayer wrote:

> Correct me if I am wrong but I am under the impression that each CPU
> has one memory controller, independent of the number of cores. So if
> you go dual core those cores are sharing a memory controller. What you
> have in the end if you go dual core with 2 CPUs is 4 cores with 2
> memory controllers.

Yes.  This is the sort of thing I am worried about.  I remember running
into problems with our Dual PIII cluster where there was a significant
difference in performance when ran we our models using 8 nodes using both
CPU's per node vs. 16 nodes using one CPU per node.  Something like 40%
faster on the single CPU per node runs. I guess the only sure-fire way to
know is to get our hands on a Dual Core system and see how our model
performs.  Anyone got one available for testing... ;^)

Steve
______________________________________________________________________
 Steve Cousins, Ocean Modeling Group    Email: cousins at umit.maine.edu
 Marine Sciences, 208 Libby Hall        http://rocky.umeoce.maine.edu
 Univ. of Maine, Orono, ME 04469        Phone: (207) 581-4302