Using 2 processors for FE analysis??? 1

[plvachon]

Hi,

Is this possible to solve a FE analysis using the 2 processors of my computer? I know the command /CONFIG,nproc exists in Ansys, but it allows the use of only "n-1" processors, in order to keep 1 processor ready for other tasks... I mention that I'm on Win XP and my computer has a dual-core processor. If anyone could help me! Thanks.

 

[Spirit]

plvachon,


many FE solvers include parallel algorithms for improved solution, usually the relevant license is not anyway included in the standard package and requires an additional expense.
Also consider that on many solvers the CPU time is not the bottleneck, the main delay being in swapping on disk, so the benefit of parallelization may not be so scalable.

Cheers


Spirit


'Ability is 10% inspiration and 90% perspiration.'

 

[Stringmaker]

Ansys recommends that you use N-1 processors. They don't say that you are held to that however. If you have a dual core processor and the appropriate license then by all means set NPROC=2. However, other applications running simultaneously may suffer performance wise so unless you envision long solve time or have no immediate use for your machine I'd stick with the recommended NPROC.

 

[rob768]

I'd suggest a test. Solve a (large) problem, time it, and then perform the same with both processors together. See if the advantage is significant enough to "sacrifice" the second processor.

 

[cbrn]

Hi,
it is still a big question: in Ansys, is the use of 2-cores processor recommended or not (i.e., in the Bios, should one set Hyperthreading or not)??? Well, as far as I understood it, the response is not univoque: certain configurations get better performance with Hyperthreading ON, thus requiring NPROC,<number-of-cores>, while some others work better with HyperT OFF and NPROC,<number-of-physical-processors>. Also consider that not all the parts of the Ansys solvers are multi-processor (though they are multi-thread). For example, if you have HyperT ON on a single physical processor machine (it's my case) and you set NPROC,2, you will see that the PCG preconditioner is single-processor (only one core will be used, thus the total processor charge will reach at most 50% + Windows activity), while the solver is able to spread through the 2 cores (total processor charge will reach 100% if the iteration keeps in memory and doesn't swap to file). At least, that's what happens on my machine... But informatics is sometimes random...

Regards