CFD Online Discussion Forums

CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)
-   OpenFOAM (https://www.cfd-online.com/Forums/openfoam/)
-   -   What hardware are you using? Tips for new user? (https://www.cfd-online.com/Forums/openfoam/114689-what-hardware-you-using-tips-new-user.html)

greenleader March 15, 2013 09:55

What hardware are you using? Tips for new user?
 
There is not not much discussion about hardware on the OpenFOAM forum :(

I am currently playing with OF in a virtual machine under XP on an oldish Intel Core-duo E6300, but I plan to move to new hardware soon.

I don't know a lot about computer hardware - my questions are:
- should I go for a single or dual processor system? (can I run on one processor and then add another later if I wanted more power?)
- what processors are recommended?
- are there any brands/processors etc that one should avoid for compatibility reasons?
- whats important when buying RAM (apart from the size)? I think I'd start with 16GB but select a system that I could expand to say 32GB if I wanted to later.

I am starting out with 2D problems with less than 100,000 cells but when I build up to 3D geometry I expect 5M cells is likely.

regards
Steve

elvis March 15, 2013 14:12

Hi,
the Hardware forum at cfd-online is not bad and there are also people discussing
on hardware that is suitable for OF, but the discussion for other software can be read as well
because the ratio 'cells/CPU" is important. I think you do not see a acceleration below 200000 cells/CPU.
Intel CPU's have better performance than AMD in this area. Some people & companies have worked on GPU acceleration for OF. You find announcements in this forum and presentations @ OF-workshop

greenleader March 16, 2013 09:42

thanks for pointing that out - I was focused on the openFoam forum and I had not looked in the more general forum. I have now found some useful tips...

kwardle March 19, 2013 12:04

I would disagree somewhat with what Elvis has said regarding cells/CPU. While it is very dependent on the physics you are looking at, for interFoam based VOF solvers I have found you can get good speedup at much higher per processor distributions (~25K cells/proc). Maybe you had an extra "0"? That has been my experience anyway. And if you are very impatient and not computational resource limited, I sometimes run even lower (7500K cells/core), but this is very aggressive and about the limit where you don't get much more speed-up and start to drop below 75% parallel efficiency. This is for multiphaseEulerFoam anyway. Just my 2 cents.

immortality August 1, 2013 08:21

hi
What do you mean by acceleration originated by "cells/CPU"?could anyone clarify more in this relation?


All times are GMT -4. The time now is 16:57.