CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Hardware (http://www.cfd-online.com/Forums/hardware/)
-   -   hardware requirements for 31m cell case (http://www.cfd-online.com/Forums/hardware/93618-hardware-requirements-31m-cell-case.html)

petrile_83 October 20, 2011 12:18

hardware requirements for 31m cell case
 
I'm working in bus industry and I'm intrested in simulating interior cooling and heating system. I found this example
HTML Code:

http://web.student.chalmers.se/groups/ofw5/Presentations/OliverWebelSlidesOFW5.pdf
What kind of hardware would be suitable for solving this kind on 31m cell case in resonable time like 80-100h? Report says they used 64 cores to solve in 88 hours. Report has no information of cpu:s they used. With kind of cluster would be needed?

abdul099 October 20, 2011 23:39

The time needed to solve depends on too much variables to say how many CPU's you would need to solve in a specific time. It depends on how many iterations you need, if it's a transient or stationary simulation etc. and also on the used software.
But for sure the CPU's and / or interconnect used in this case are quite slow.

Just an example from two clusters I've been working on:

Running a 120M cells for 5000 iterations case takes about 20h on 256 cores (CPU's based on Intel's Nehalem architecture, frequency not known) with Star-CCM+.
Same software, 97M cells, 64 cores, 6000 iterations took about 3,5 days. CPU's: Intel Xeon X5550, 2,66GHz.

Both clusters have been some "common" Dell clusters. Nothing special but also nothing build with computers purchased at Walmart.

rwryne October 21, 2011 11:57

Quote:

Originally Posted by abdul099 (Post 328834)
The time needed to solve depends on too much variables to say how many CPU's you would need to solve in a specific time. It depends on how many iterations you need, if it's a transient or stationary simulation etc. and also on the used software.
But for sure the CPU's and / or interconnect used in this case are quite slow.

Just an example from two clusters I've been working on:

Running a 120M cells for 5000 iterations case takes about 20h on 256 cores (CPU's based on Intel's Nehalem architecture, frequency not known) with Star-CCM+.
Same software, 97M cells, 64 cores, 6000 iterations took about 3,5 days. CPU's: Intel Xeon X5550, 2,66GHz.

Both clusters have been some "common" Dell clusters. Nothing special but also nothing build with computers purchased at Walmart.


What kind of simulatinos are you doing, if I may ask? Kind meaning what physically are you simulating

abdul099 October 22, 2011 21:55

Quite different cases. Could be multiphase, external aero, pumps, wind turbines, CHT...
So far no chemical reaction, no combustion, no LES / DES. For sure there are some other things still missing I didn't mention here.

The cases I mentioned in my previous post have been an external aero case on a racing car and a pump. Both with standard physics: Incompressible, constant density, isothermal, RKE turbulence...


All times are GMT -4. The time now is 19:00.