WHICH HARDWARE ???

[LOFO]

Hello. I need to know if one of you can help me selecting the best up to date hardware for intensive finite element analysis on CATIAV5. I am in the composite molding business. Most parts we produce do not have a single straight line and are of orthotropic material nature. It takes quite a powerfull machine to perform such analysis in a reasonable time... It has to be fast, fast, fast regardless the budget for a PC.
I need help from a pro
Louis

 

[MToft]

Hi,
I tend to use Abaqus or Nastran but requirements can be pretty similar - 2GHz+, 4Gb RAM, 100Gb memory seem to failry standard minimum levels these days. One trick that helps me is keeping the hard disk reasonably clear and defragmented - if the program can write its temporary files in one solid block of empty memory then it wastes less time jumping between locations, which seems to waste LOADS of time. For example, reducing hard drive usage below 50% and freeing blocks above 10Gb has accelerated analysis for me 20-30%. Have you asked CATIA what they recommend?
Good luck...
MToft

 

[ThomasH]

Are you talking about CATIA as a solver (do they have one?) or is CATIA your "drawing" tool?

What kind of solver do you use? You have to realize that a complex geometry does not directly influence the solution time, only the size of the problem. There usually is a connection between complex geometry and many DOF's but it does not have to be that way. The easy answer is to buy the most expensive PC (?) you can but that can be a bad idea depending on how the solver is written, parallell code etc.

Also, what type of computer do you have today and what is the normal solution time. "Fast" can be anything from fractions of seconsd to hours or days.

Regards

Thomas

 

[LOFO]

Hello again.
About best possible hardware... I am basically modeling on CATIAV5 then I run analysis with the Structural Analysis module, Elfini solver. Also and the slowest to run is the HBR or human modeling solution that makes manikin with full degree of freedom at all joints. So imagine when I run posture analysis using a "shutle" and trajectory with DMU kinematics it takes forever... and it crash many times.
Looks like a twin Xeon would do the trick but basically it takes computing power not graphic performance.

Louis

 

[TGS4]

If your problems are particularly large and will not solve in-core (that is, in the physical RAM) I would highly recommend that you abandon your 32-bit operating system and computer and move to a 64-bit system. Then, you will be able to have (for a reasonable price) up to 16GB RAM (or more, depending on the actual architecture).

For a PC (something that could eventually run Windows - whenever they _finally_ release XP-64), your processor options are AMD Opteron and Intel Xeon. For my money, I have gone with the AMD - the benchmarks appear to be superior and the motherboards allow 16GB RAM.

OTOH, if your problems aren't particularly large and run in-core with 3GB RAM or less, then you will simply need more processors. You should be able to get quad processor computers from most computer companies. If that isn't enough, then you should look into clustering technologies, and create your own "super"computer in a cluster environment.

 

[ThomasH]

LOFO

Before you go and buy a twin Xeon machine you shold make sure that Catia utilizes it's features. There are solvers that actually run slower on Xeon than pure P4 technology. The solver must be written for Xeon (typically machine code) to be really efficient.

I have a twin Xeon so I know that for a fact.

So far you have described your problems as "big" and your requirements as "fast". I don't mean to be rude but it does not say a lot. Are you refering to millions or even billions of DOFs?

A different way to go would be to connect Catia to a industri standard solver, they usually beat the competition. But I'm not familiar with the Elfini solver.

Regards

Thomas

 

[amorrison]

Post your question also to

http://www.cfd-online.com

Also search their question archives.