How many cores is too many?

Discussion created by Guest on Sep 29, 2006
Latest reply on May 11, 2007 by Francesco Rustichelli
Now that cosmos can work with multiple processors, how many is good enough?
My analysis I'm running today has 930,000 DOF, 11 bodies total, 5 bonded surfaces, 5 contact surfaces, and 4 bolted surfaces each held with 12 bolts. 6 bodies are steel, 5 are nylon, so I am running direct sparse solver.
This is a big simulation and is at 35% after 2 days running on dual core machine with Windows XP Pro (32-bit), 3GB ram, two 3.4GHz Xeon chips.
I'm going to get a new computer because I will have lots of analyses more or less about this size. The IT people are considering Windows XP-64 bit, 8GB or 16GB of Ram, and perhaps 4 or even 16 processor cores. Then my neighbor mentioned that FEA is notorious for suffering from diminishing returns from multiprocessing, that perhaps 4 cores is already too much and 16 cores is likely too much. Anybody have any background on this topic?
The question is, is 16 cores too much? 4 cores too much?
Thanks greatly for your consideration.