|
[Sponsors] |
July 7, 2011, 05:22 |
PC specification for OpenFOAM
|
#1 |
Member
Join Date: Jul 2011
Posts: 42
Rep Power: 15 |
Hi. I am planning to buy a new PC to run complex CFD simulations using OpenFOAM. What shall I be looking for when buying such machines? I would be grateful for any recommendations!!
Thank you in advance. |
|
July 7, 2011, 05:46 |
|
#2 | |
Senior Member
BastiL
Join Date: Mar 2009
Posts: 530
Rep Power: 20 |
Quote:
Regards Bastian |
||
July 7, 2011, 08:17 |
|
#3 |
Member
Join Date: Jul 2011
Posts: 42
Rep Power: 15 |
Hi Bastian.
Thank you for the quick reply. I will be looking into external subsonic/transonic compressible flows. My grid might reach 5 million cells. What would you think might be the cheapest option? Once again thank you for your time and response, Regards A.D.E |
|
July 7, 2011, 10:23 |
|
#4 |
Senior Member
Olivier
Join Date: Jun 2009
Location: France, grenoble
Posts: 272
Rep Power: 18 |
Hello,
As always, it depend of how money you will spend on ... Anyway, with a small amount of money, the best is to buy 2 to 5 cheap PC with core i5/i7 @ 3.3Ghz with 4 Go ram and Ethernet (~500€ each) connection throug a GigaByte switch. Speedup are far better with multi node than multi core. NB: i've got a far better speed up with 2 core i5 node, with a 2x2core : 4 process, than on a signle i5 with 4 process. With 4+4, speedup are not so good. NB2: with ethernet, do not use more than ~ 5 node, unless you add QDR infiniband card. Olivier |
|
July 8, 2011, 07:32 |
|
#5 |
Member
Join Date: Jul 2011
Posts: 42
Rep Power: 15 |
Hi Olivier,
Thank you for your reply. My budget will be between 2500-3000 pounds Uk I hope. So you think that it might be better to buy more than one machine and connect them together? I can not buy one PC and add cores on it?? My knowledge on setting up PCs is limited therefore I would like something good and simple.. Do you have anything like that in mind? Once again thank you for your consideration and time, Sincerely, A.D.E |
|
July 8, 2011, 08:42 |
|
#6 |
Senior Member
Olivier
Join Date: Jun 2009
Location: France, grenoble
Posts: 272
Rep Power: 18 |
Hello,
I think the main limitation come from Cache memory. 1) If you go with one multicore station, like dual xeon (i.e "professional workstation"), you may get à 2*6 core with one machine @ ~ 3000€, so 2x6 = 12 core, with 2*12 = 24 Mo Cache, and ~ 2.6 Ghz/core. 2) If go with 4 pc, core i7, you get 4x4 =16 core and 4x 8Mo = 32 Mo Cache, 3.4Ghz/core, at less than 3000€, and with real speed up. NB: at work, with the same test case, a single core i5 run on 4 core @ 3.3 Ghz are 2.5x faster than a dual xeon, which run on 8 core @ 2.2 Ghz, and with 2 station with core i5, it's 4x faster, so yes, frequency matter too.) NB2: don't forget to set on one machine a large amount of memory (8/12 Go at least) + graphic card when you will mesh/decompose/ and use paraview ... NB3: for all other station, 2-4 Go is ok if used only as a compute node (1Mio cell ~ 1 - 1.5 Go ram), and no need of hard drive/screen/... olivier |
|
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
Backflo direction specification method | vijay | FLUENT | 6 | August 27, 2015 13:29 |
Thermal specification for CHT. | mehn | STAR-CCM+ | 7 | February 18, 2012 07:47 |
Turbulence Specification Method | Srivatsan V. Rajagopalan | FLUENT | 5 | January 31, 2011 07:39 |
X,Y-velocity component specification method | Itsme | FLUENT | 0 | August 31, 2010 10:26 |
Turbulence Specification Method for pressure outle | co2 | FLUENT | 0 | April 28, 2004 20:32 |