CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > Hardware

Hardware sizing for CFD

Register Blogs Members List Search Today's Posts Mark Forums Read

Reply
 
LinkBack Thread Tools Display Modes
Old   May 13, 2009, 14:10
Default Hardware sizing for CFD
  #1
Member
 
Join Date: May 2009
Posts: 51
Rep Power: 8
gfilip is on a distinguished road
Hello,

I am currently running on 2x Intel Xeon X5450 3.0GHz quad core, with 16gb of RAM. A simple 2D turbFoam simulation (~100,000 cells), 100s, and writeInterval = 0.5s takes approximately 6 hours.
What sort of PC / cluster should I be looking at to reduce the simulation time to say 30 minutes? I understand this is a rather difficult question to answer, but I would appreciate even ballpark numbers / specs.

Thanks
gfilip is offline   Reply With Quote

Old   May 14, 2009, 16:59
Default
  #2
New Member
 
HP Kritzinger
Join Date: May 2009
Location: South Africa
Posts: 25
Rep Power: 8
mystic_cfd is on a distinguished road
Hi Greg,

some thoughts:
my experience is with fluent and cd-adapco products mostly. don't know how openfoam scales.

CPU power:
Using Fluent benchmarks a X5570 Nehalem should be 45% faster than your 3Ghz Harpertown (per core). In addition, we see speedups on dual CPU quad-core Harptertown box of around 4.3 (on all 8 cores). Benchmarks show this to be in the range 6.5-7.5 for Nehalem. So the same box with Nehalem quad-cores could give you a speed-up of ~(1.45 x (7/4.3)) = 2.4. That only gets you down to 2.5hrs on your solve time...

problem size:
In our experience we start seeing "negative" speed-up when going below 20000 cells per parallel partition. Using smaller cell blocks starts to increase solution time due to the large amount of communication required between partitions. So for 100000 cells you might already be hitting that limit - before buying anything, first check this effect on existing hardware (your current setup might give faster results if you use 4 partitions and run only 4 of the cores in your 8 core box - but take care to do core pinning so that each process gets its own CPU cache).

disk write speed:
if you write excessively, you might be waiting for disk writes instead of the solver. You must check this before buying anything. An option would be to run a RAM drive where you store output data during runs. (writing solution monitors and outputs can have the same effect - on very small problems I have seen the solver wait for the terminal or file - turn off or reduce the frequency of writing residuals etc. and check if this is a problem on your setup).

regards
mystic_cfd is offline   Reply With Quote

Old   May 15, 2009, 15:36
Default
  #3
Member
 
Join Date: May 2009
Posts: 51
Rep Power: 8
gfilip is on a distinguished road
thank you very much for this valuable information. i will have to look into several issues you presented.
gfilip is offline   Reply With Quote

Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
Hardware recommendation? AMD X2, Phenom, Core2Duo, Quadcore? rparks OpenFOAM 0 April 22, 2009 09:10
Gambit: interaction of two sizing functions? Ralf Schmidt FLUENT 3 January 30, 2006 18:35
Error with sizing function Ralf Schmidt FLUENT 1 October 31, 2005 11:59
AMD opteron hardware gg FLUENT 4 September 9, 2004 11:57
hardware Guus Jacobs Main CFD Forum 24 May 3, 2000 16:42


All times are GMT -4. The time now is 17:27.