A good set of powerpoint Notes on CFD simulation

[ConnClark]

This is a presentation on using OpenFoam in a commercial setting. It does however contain information that is applicable to to all simulations on cars such as how simulation of ground motion effects things and what sort of computing power you need to get reasonable results (within 1% error).

http://www.opensourcecfd.com/confere...2009_Islam.pdf

[ChrstphrR]

Nice!

So, if I ever do get around to making a decent model, and figure out openFoam to run it in parallel on my two computers... scalling for the processors, in number, and performance, to model a car of the same complexity they do at Audi, it'll take me about 2 months what they can do in 25h on a (fairly large) cluster...

:|

[ConnClark]

On the CPU horsepower side it may not be quite so bad because clusters don't scale linearly. On the other hand you will want/need lots of RAM and with out it it will take much more than 2 months.

[ChrstphrR]

Originally Posted by ConnClark

On the CPU horsepower side it may not be quite so bad because clusters don't scale linearly. On the other hand you will want/need lots of RAM and with out it it will take much more than 2 months.

I'd kept your recommendations in mind for quite some time now, from back in this thread:

http://ecomodder.com/forum/showthrea...tml#post215720

Currently, though it will change when I get time, my desktop runs XP, 3gb, and a Core2Quad (2009). The other machine uses Ubuntu, and the Core2Duo that was in the other machine originally, and 8Gb ram.

I'm contemplating swapping the processors around, or finding another Core2Quad to fit in the second machine. Eventually I'll make the former machine dual boot, or run XP on a VM, and then I'll try to get 8 or 16gb of RAM installed.

I'm still quite interested in dcb's thread about "OpenFoam for the masses", just have not had time nor resources to put into the logistics of it all... until now.

[ConnClark]

ChrstphrR,

Glad to hear your still playing with stuff. I haven't had the time.

I'll give you some more rule of thumb information. Going from one core to two cores gives about a 60% boost in speed running CFD code. Adding a third core gets you about a 25% boost over that. Adding a fourth core gets you about 10% again over that. The reason for this drop is the cores compete against each other for memory access. Larger CPU caches help a little. Faster lower latency memory helps more.

If your simulation needs more memory than you have and starts to swap to disk its best to have one swap file located on each physical disk in the system instead of one big swap file. A permanent swap partition is marginally faster than one created in your root directory.

As far as setting up openfoam for the masses there is a project that has done so.

Greenpower VWT - Greenpower VWT - Virtual Wind Tunnel

Greenpower VWT

They state that their results aren't very accurate but that is what you get when you do CFD with a dumbed down interface.

Good luck

[Vekke]

Here is also some cfd data and results:
Automotive aerodynamics-optimization---2013-07-17