This content has been marked as final. Show 6 replies
On a quad core, wouldn't full utilization be 25% on 4 cores, 50% on a dual core, 100% on a single core?
have you set the processor cores affinity manually for the solver process? That may help.
How do you "set the processor cores affinity manually for solver process"?
I have observed the same
I am using similar at home
AMD 9500 quad core 2.2GHz
4 Gig Ram
500 G hard drive
ATI Radeon 3850 512 mB direct x 10 graphics card (seems ok for home use but no real view)
Vista 64 bit
Processor affinity simply limits processes to run on the core(s) you specify. I wouldn't imagine there is any benefit to changing it unless you are doing other tasks in the background and want to have some spare processing power left over for that.
if you REALLY want to play with it, ctrl-alt-delete and bring up the task manager. Then go to the processes tab, select the cosmos solver process, right click, and choose Set Affinity... Youll be able to choose which cores it is allowed to run on.
At any rate, the vast majority of Solidworks code (including a lot of cosmos) is not optimized for parallel processing. In these cases, you'll only see the software utlizing one core. For a quad core, Windows will report this as 25%.
i see a similar problem on a 8-core mac (running vista 64)
although i get a better performance: 60% for the iterative solver and 50-60% for solving contact restraints..but for 'bout half of the time the load is 1% and i wonder what cosmosworks is chewing on during this time
the AMd CPU's are not officially supported or at least they weren't a release or so ago which I complained about quite a bit.
It depends what you are doing.
-Meshing is a single process operation so no joy on that.
-Matrix solving is vectorized and sparse solves does this pretty well as long as you have the resources.
-Contact iterations - single process only so 25% is all you will get on this task as best as I can figure out.
-Non-linear - the matrix inversion are vectorized but he convergence iteration are not as best as I can figure out.
hope this helps