CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Hardware (http://www.cfd-online.com/Forums/hardware/)
-   -   Memory of PC for running fine mesh size (http://www.cfd-online.com/Forums/hardware/90224-memory-pc-running-fine-mesh-size.html)

sepidecent July 5, 2011 02:44

Memory of PC for running fine mesh size
 
Hello

Im running simulation with 8 million elements and more (maybe 10 million),the current PC with 12 GB Memory is not able to do the simulation because of lack of memory and I'm going to buy a new Dell desktop computer with 96GB(6*16GB) DDR3 RDIMM MEmory,1066MHz,ECC-but I'm not too sure if the new one is able to handle the simulation or not.I was wondering if any body could help me with the memory of PC I need for simulation.

Regards,
Sepideh

abdul099 August 7, 2011 16:17

It depends on the software you are using and the complexity of your case.

For star-ccm+, as a rule of thumb, 1 million cells should use about 1GB memory on a standard case. So I recommend to monitor the memory usage when running one of your cases. Maybe there is just something else which uses too much memory (like operating system, CAD, firefox or whatever).

Anyway, 96GB should be much more than needed to run a 10 million cells case. On my 48GB workstation, I already run cases with up to 50 million cells (with star-ccm+).

sepidecent August 7, 2011 18:50

Hi

Thank you for comment,I am using CFX12.

stuart23 August 16, 2011 07:48

Hi Sepideh,

Although there are rules of thumb out there to relate memory to mesh size, you must also remember that the each node is only an element of your entire simulation. If your simulation is simple, i.e. incompressible laminar flow, you can probably have a much finer mesh than if your simulation is more complex. Using LES, Reynolds Stress, multiphase etc etc will all add extra variables that need to be stored somewhere, and will therefore need more memory.

In my experience, I have done 80M element LES simulations and they have used over 200GB of memory. Using RANS models, they only used 40-50GB though.

Also, if you are solving large simulations on a single or dual socket computer, do not expect fast convergence!!! There are some very high end multi socket server boards out there as well, but the Quick Pass Interconnect may not be able to keep up with all the processors. Have you considered parallelisation?

Stu

sepidecent August 16, 2011 20:43

Hi stuart23

Thanks for reply.

Cheers,
Sepideh


All times are GMT -4. The time now is 08:26.