CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   CFX (http://www.cfd-online.com/Forums/cfx/)
-   -   Core usage on CFX parallel processing (http://www.cfd-online.com/Forums/cfx/26953-core-usage-cfx-parallel-processing.html)

alterego January 29, 2009 09:33

Core usage on CFX parallel processing
 
I did some calculations on our new server with two Xeon X5450 Quadcore CPUs. Due to our license we're allowed to run on 4 cores.

As I had a look on the system monitoring I spotted that the four parallel threads were running only on one cpu. To avoid cpu architecture limitations I want to run those 4 parallel processes shared equally on both CPUs- e.g. 2 processes on two cores of CPU1 and 2 processes on two cores of CPU2.

Any suggestions?

I already tried running on PVM, HP MPI and MPICH (all local parallel).

PS: We're running on Suse Linux 10.

Beginner January 29, 2009 11:39

Re: Core usage on CFX parallel processing
 
Hi there,

I have a same problem. I want to run parallel solver on windows 64bit system. I already tried running on PVM, HP MPI and MPICH (all local parallel).

But there is always this error message:

A CFX-5 Partitioner executable is required for this run, but none is available.

Can anybody help me?

Thanks

Beginner

CycLone January 29, 2009 12:35

Re: Core usage on CFX parallel processing
 
Did you register MPICH in Windows? If not, follow the installation instructions for MPICH on Windows provided with CFX. HP MPI isn't available in 11, but you can try that if you have the v12 preview release.

-CycLone

CycLone January 29, 2009 12:40

Re: Core usage on CFX parallel processing
 
The OS controls where the processes run. Under Windows you can set the affinity (i.e. what processor to run on) in the Windows task manager, though you'll need to figure out which cpu is on which processor. You'll also need to figure this out on Linux (don't just assume CPU 1-4 is on 1 and 5-8 is on the other) and there are similar utilities to assign the affinity.

HP-MPI generally does a good job of balancing it out. In addition to balancing processes between processors, you need to split the processes between cores that are served by different caches (there are two L2 caches on each Intel chip which serve 2 cores each). Again, HP-MPI usually does a good job of this, but it also depends on what else you are running on your system.

-CycLone

cfdguy January 29, 2009 12:52

Re: Core usage on CFX parallel processing
 
Your issue concerns the new Intel CPU released after CFX11. Contact your local support and they will provide a single patch for that.

I've been through this before. Good luck!

lerr December 21, 2011 03:38

Is it possible to find this patches on-line?

ghorrocks December 21, 2011 06:45

The ANSYS Customer portal. But if you have active TECS you should ignore this old problem and just go straight to V14. Zillions of bugs have been fixed from V11, and lots of new features.


All times are GMT -4. The time now is 14:29.