SMP Scalling Performance
Hi All i'm currently running SMP with 4 CPUs and would like to understand if any user out there tried >4 CPUs with SMP version for actual application? and what is the scaling performance?
|
Re: SMP Scalling Performance
Yes, and have seen up to 3.5X speed up with 8 CPU's.
|
Re: SMP Scalling Performance
Melissa have you tested with actual application? what kind of physic is involved?
|
Re: SMP Scalling Performance
The performance of the SMP code varies depending on the models activated. We've benchmarked FLOW-3D on a dual/quad Intel Harpertown 2.8 GHz Xeon 5462 w/1600 FSB using 8 processors. All simulations continued scaling out to 8 processors and the scaling ranged from 1.3x for an axi-symmetric noozle simulation with about 150,000 cells to 3.4x for a shallow water simulation of a reservoir.
Other simulations run on 8 processors: Fuel Sloshing, isothermal, GMRES pressure solver, 2 million cells: 3.1x High Pressure Die Casting, explicit heat transfer, GMRES pressure solver: 2.85x Lost Foam Casting, implicit heat transfer, 250,000 cells, 2.63x |
Re: SMP Scalling Performance
John Thanks for inputs and feedback. But have you tried it with GMO + turbulent model? since GMO is program to run in serial so would like to understand if any benefit to run with 8 cores SMP. Thanks.
|
Re: SMP Scalling Performance
The GMO model is not parallelized at this time so you will not see a significant speedup when GMO is active. Turbulence will not affect the scaling however.
|
Re: SMP Scalling Performance
John Thanks for the inputs and btw can you send me your SMP case study report for reference?
|
Re: SMP Scalling Performance
Sure, send me an email at john@flow3d.com and I'll send it to you.
|
All times are GMT -4. The time now is 09:29. |