|
[Sponsors] |
December 17, 2007, 17:42 |
Help: Computers for large scale simulation
|
#1 |
Guest
Posts: n/a
|
Dear all,
The company I work for has decided to upgrade my computer and are currently looking into new machines from Dell. Does anyone have any recommendations? Thanks a lot for your information. sincerely, sarah |
|
December 18, 2007, 07:31 |
Re: Help: Computers for large scale simulation
|
#2 |
Guest
Posts: n/a
|
We just ordered two new pcs from dell specifically for CFD work. Spec obviously depends on your cfd requirements, parallel licenses and budget available.
Well make sure its a 64 bit system (so that there's no limit to the RAM accesed by the cpu) and that the OS is also 64 bit (sounds obvious i know, but you'd be suprised..it can be an issue esp when trying to integrate the pcs to a company network) It's also worth checking with your cfd provider for recomnended hardware and benchmark cases to get a feel of what you actually need. |
|
December 18, 2007, 08:44 |
Re: Help: Computers for large scale simulation
|
#3 |
Guest
Posts: n/a
|
If you have windows based machine then go for 64 Bit, dual-core dual cpus with 16 GB RAM and each 2.2 GHz+ and 64Bit XP Professional version. This is what I have and it works very well.
|
|
December 18, 2007, 16:34 |
Re: Help: Computers for large scale simulation
|
#4 |
Guest
Posts: n/a
|
Thank you all for your kind help!
|
|
December 19, 2007, 05:28 |
Re: Help: Computers for large scale simulation
|
#5 |
Guest
Posts: n/a
|
1. Don't go for windows (we run redhat entreprise 4 WS on our Dell cluster). Linux is more stable, less maintenance, less overhead. 2. So far it looks like an intel Xeon 5160 is still the fastest for CFD (or you have to wait a couple of months for a new faster dual core)
|
|
December 19, 2007, 07:23 |
Re: Help: Computers for large scale simulation
|
#6 |
Guest
Posts: n/a
|
We have both UNIX and windows over here. I need windows because the automatic meshing from CATIA to FC is only supported in windows as of now. Still never had any issues and no one from the finance department is complaining !!!
|
|
December 20, 2007, 05:40 |
Re: Help: Computers for large scale simulation
|
#7 |
Guest
Posts: n/a
|
Most CFD codes scale with integer speed. Benchmarks are regularly updated here:
http://www.ideasinternational.com/be...0-fc32c71b1dd0 |
|
December 20, 2007, 06:04 |
Re: Help: Computers for large scale simulation
|
#8 |
Guest
Posts: n/a
|
I thought CFD was a floating point thing. Graphics is more integer related. And you want the best performance per core for (parallel) CFD which is not necessarily reflected in these benchmarks. A quad core intel might be faster per CPU but not per core. As parallel CFD jobs run on individual cores and are limited also by memory access bandwidth, multi-core processors are not the best options for CFD. If you really want a fast CFD cluster you can use a stack of Playstation 3 computers(vectorized computers) see: http://www.linvision.com/
Is also recommended for some nice R&R in between CFD runs ;-) |
|
December 20, 2007, 06:20 |
Re: Help: Computers for large scale simulation
|
#9 |
Guest
Posts: n/a
|
A better reference on PS3 and scientific computing:
http://www.netlib.org/utk/people/Jac...PERS/scop3.pdf |
|
December 20, 2007, 10:17 |
Re: Help: Computers for large scale simulation
|
#10 |
Guest
Posts: n/a
|
Most CFD codes are held up by indirect addressing, which is integer-based. Direct solvers (prevalent in FEA codes) do matrix inversion and other things that LAPACK et al have optimised over the years - they go with floating point speed. Where I work we write both types of code and can see the clear bias towards different architectures for the two.
|
|
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
Axial fan simulation in UG/NX 7.5 | fan123 | Main CFD Forum | 2 | April 23, 2011 09:22 |
Continuous vs interrupted simulation | sega | OpenFOAM Running, Solving & CFD | 4 | November 3, 2008 15:29 |
Overflow problem in steady simulation | ReeKo | CFX | 11 | October 8, 2008 18:57 |
Large Eddy simulation | Andreas Hauser | Main CFD Forum | 1 | May 20, 2000 21:33 |
3-D Contaminant Dispersal Simulation | Apple L S Chan | Main CFD Forum | 1 | December 23, 1998 11:06 |