More on solver speed: case study

[mastroit2002]

Dear Sirs,

I'm looking for an advice: is there a list of structural linear case study usable to test commercial FEM packages, like "driven cavity" or "facing step" for CFD codes.

Another question: is solver speed cost effective in Your own activity?

Is there anyone using a dedicated cluster for nonlinear structural dynamics?

Regards,
Mastro

 

[GBor]

I run non-linear structural dynamics on a dual pentium workstation using Algor FEA software. I do impact problems, but not automotive crash analyses...most of mine are hammer strikes, dropped objects, etc. comparatively low-speed. When you get to that level of analysis, solver speed is definitely cost-effective. Most software packages offer their own "Accuracy Examples" that you can model yourself. They show you the calculations which are generally out of some recognized, referenced text that you can verify. NAFEMS also has some standards, but you may have to understand the nuiances of your software to make certain you are inputting the proper values and element types.

Garland E. Borowski, PE
Borowski Engineering & Analytical Services, Inc.

http://www.borowskiengineering.com

 

[cbrn]

Hi,
in my company instead solver speed is not of primary importance, because we have lots of automated routines which allow batch processing overnight. There is one exception, though: the CFD department runs on a dedicated cluster.
It's more a question of memory: with 1 GB, I am limited in the size of the problem I can handle: it's a "pass-through" criterion: if the solver CAN start, no matter (well, approximatively of course...) how much time it will take it will be much faster than testing a prototype and much cheaper than to afford the risk of a failure... But if it CAN'T, then I may have the fastest processor in the world but it will serve to nothing at all...

Regards

 

[GregLocock]

We run linear and non linear FEA, and CFD, on dedicated Crays, but we also run some jobs locally. I often run dynamics jobs on many cpus (up to 8), but as separate jobs, not multithreaded or anything odd.

The CFD guys run multithreaded jobs locally, I don't know any details.

Solve time is a big issue - if my models would run in 10 seconds instead of half an hour then I'd be far more productive, as I wouldn't be reading eng-tips, or get otherwise distracted!, between runs.





Cheers

Greg Locock

Please see FAQ731-376 for tips on how to make the best use of Eng-Tips.