CFD Online Discussion Forums

CFD Online Discussion Forums (
-   Hardware (
-   -   Home Workstation for CFD (

snowculli April 30, 2012 11:34

Home Workstation for CFD
Hello to everyone,

I'm trying to built a home workstation for my CFD final project for university and planing to use it in motorsport industry. I'm creating my own aerodynamics consultancy at the moment for small motorsport teams using OpenFoam software.

I've found this component configuration:

2 X AMD Opteron 6212 8 Core a 2,6GHz 16 mb cache Motherboard Asus KGPE-D16/IKVM - 2 G34 socket
16 x Kingston HyperX DDR3 1600MHz 2GB RAM ATI FirePro V4800 (100-505606) 1024MB
What do you think about this workstation? the main problem is the budget, so i don't want to spend more than 2000 eu... with this configuration i just arrive to 1600 eu.

What about the AMD Opteron 6212?

But i'm thinking to buy four socket Motherboard for future cpu adds... but it's 400 expensive.... what do you think as my first workstation?

Do you have any suggestions?

Thanks a lot for your help!! best regards!

- Albert

abdul099 April 30, 2012 19:11

Well, 32GB memory is not that much. F1 cars I used to run had much more than 100 Million cells, therefore you would need much more memory to run that.
That's why I think, it's not worth to go for more than 1 cpu. You will most probably never run a serious motor sports car model on a workstation. Not even with 4x 8 cores, just because the memory bandwidth will be crap. It might be better to save the money for every additional cpu but the first one and go for a computing service like Amazon.

You should also consider, AMD cpu's are not very fast for CFD calculations. I have a Phenom II quadcore running on 3,4 GHz, and it's about half the speed of the old and slow Xeon X5550 2,67GHz I have at work.

So although it might be more expensive, I would go for a fast socket 2011 Sandy Bridge (some motherboards can handle up to 64GB memory).

snowculli April 30, 2012 20:41

it's right that for runing a 100 milion cells simulation 32 gb of ram isn't enough. I've been doing an aerodynamic and CFD job in a motorsport team and we were working with 6 milion cell mesh (optimized, gradual reduction of cell size as you getting closer to the model) for all the whole car, and compering with the track test results we just see that the results are not far enough from reality.

The workstation that i want is for using to support small motorsport teams to help improving their cars.

What about overcloking the CPU's? will t give me any advantatge?


CapSizer May 2, 2012 05:38


Originally Posted by snowculli (Post 358551)

What about overcloking the CPU's? will t give me any advantatge?


As far as I know the multi-socket motherboards don't normally have the kind of overclocking controls you find on enthusiast's single socket boards. The evidence seems to suggest that the most important single parameter to focus on is the memory. Have you checked that the board will support the memory that you have chosen? I think multi-socket systems need ECC memory.

Also, and this is the difficult part, you need to try and compare the performance to a similarly configured Xeon E5 system using equivalently priced CPU's. The Xeon will not have the same number of processor cores, but the faster per core performance may make up for that. The 2.4 GHz E5-2609 quad-core processor costs more or less the same, but has half the number of cores. The published SpecFPrate results indicate that the two systems perform essentially the same, but how well that translates to CFD performance may be a different question. If you are going to be using commercial software that is licensed per core, the Xeon will be far more cost effective.

snowculli May 2, 2012 08:46

The motherboard choosen is this one:

Four Sockets G34 with a non ECC ram up to 128 gb, and ECC ram up to 1Tb. Thanks for the advice i didn't chek it before.

And about the Processor an AMD Opteron 6212:

Isn't bad for that price... 266$.

I'm goingo to use OpenSource software... The license cost of commercial softwares like Star-CMM+ are so high for my starting budget... And i prefer to understand how to write my own CFD code it's more challenging.

How about the Video Card? What do you recomend a Engineering video card like
ATI FirePro V4800 1024MB 180$ or a game video card for the same price?
I will use the worksation also for a CAD solutions using Catia, NX7.5 .....

Thanks again!

CapSizer May 2, 2012 09:46

OK, well if you are going 4-socket, then the Xeon E5-2609 is no longer a direct comparison. On the face of it, a dual E5-2680 might be comparable, but unfortunately the SpecFPrate numbers for these make use of the hyperthreading. From what I've heard, this does not work so well for CFD, and in any event, these CPU's are massively more expensive. So I reckon your choice of system makes quite a bit of sense, but double-check the question of the non-ECC memory (speak to somebody at Supermicro), it is very unusual.

Somebody else with advice on the graphics card? I would also like to know how much of an advantage you get from the professional graphics card.

Also, it may take a bit of searching to find a good case for that motherboard. it is LARGE. It is reasonably easy to find chassis for EATX workstation boards, but the SWTX boards are really made for rack-mounting. See

abdul099 May 2, 2012 16:32

Hm, it seems I massively overestimated the size of your typical cases. For sure you can run a 6M cells mesh on a workstation. But I still think, 6M cells is a very very low number.

When choosing a CPU, don't look only on the floating point performance. It's better to have a look on the overall performance of a typical system. FP performance is often evaluated by using synthetic benchmarks without properly taking memory bandwith and communication latency into account - which is very important for CFD applications.

Regarding the graphics cards, I think, gaming cards will offer more performance. BUT many commercial applications like CAD don't properly support them. So although it is more important, I would go for an engineering card. CAD shouldn't need a extremely powerful card. Most cards should offer much more performance than you need (from the hardware side). Some software might not use the hardware in an efficient way (ccm+ is an example), but in this case, it will not help to go for a faster card.

JBeilke May 2, 2012 17:41

If you read the README for some of the CFD related softwares you will find, that an NVIDIA Quadro is recomended. A Quadro 600 is usually enough.

All times are GMT -4. The time now is 11:59.