Multi-treaded finite elemental analysis software?

[voxeltree]

Is there an efficient and cost effective a Multi-treaded or Multi-CPU Post processing FEA software package available? I need something that will utilize all of my 8 core processors, to speed things up?

 

[FeX32]

Check out Comcol Multiphysics. Great package, and I do believe it supports multithreading.


Fe

 

[voxeltree]

Okay thanks, I'll check into that. I did find out that Abaqus is multitreaded, but im not sure if they charge you more for it or not. I'm not sure what FEA software to use. It's so nebulus!

"The distributed memory parallel direct solver
technology in Abaqus/Standard sets a new benchmark
for the industry, demonstrating impressive scaling and
robustness. Today, effective parallel scaling is available
on up to 64 CPUs for a range of industrial applications;
models that may take days to run on a single CPU can
now be run overnight."

 

[DanStro]

For Abaqus I don't remember the what the extra costs are but the price does increase based on number of cores that you use.

 

[voxeltree]

Do you know anything about Nastran?

 

[ThomasH]

I thing MSC.Software charge extra for multithreading. NEi.Software has a 64 bit solver that utilizes several cores and the price is not to high. As for NX Nastran, I don't know.

Regards

Thomas

 

[GregLocock]

I'm pretty sure MSC Nastran and LS Dyna can be multithreaded. Using 8 cores your solve time should be improved by 60% or so over two cores (depends on how good the partitioning is). They charge additional license fees per cpu.





Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[TERIO]

What kind / size of problems are you trying to model? Also, be aware that if your problems are more moderately sized splitting it up between CPUs may not have much (or any) benefit because the communication between them is not free.

 

[ThomasH]

I would be very surprised if LS Dyna wasn't multithreaded since it is explicit. However, there is usually an additional cost.

Regards

Thomas

 

[ESPcomposites]

What is odd is that you are concerned about speed, but don't really care about what software you use?

Using the right software (tool) for the job and becoming very familiar with it seems more important than multi-threading. Besides 95% of your time is learning your tool and getting to work right, not solving.

What software are you familiar with?

Brian

http://www.espcomposites.com

 

[GregLocock]

Brian a full crash sim can take a week or more to solve. In the past I've run DOEs and searching algorithms using automatically modified models that run for weeks as they step through different iterations. So sometimes speed does matter.

That being said in the latter two cases a bit more thought in the design of the experiment could probably eliminate many of those runs.



Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[voxeltree]

Thanks for all of your help. Most computers run on multiple CPUs now, so why not use an app that allows you to exponetially increase your post processing power. MSC Nastran has solvers for parallel and multithreaded processing that allows users to take full advantage of their multi-core systems without the added cost of other applications. NASTRAN is an FEA program that was originally developed for NASA about 30 years ago for the digital prototyping process. I love the user interface and it has a lot of meshing options. I'll let you know how my training goes.

 

[voxeltree]

http://www.mscsoftware.com/events/Webcasts/MDNastran/WhatsNew2010/

 

[ESPcomposites]

Greg,

Well this is certainly possible and I have had a few runs that brought Lockheed to its knees, haha. But in hindsight, as you mentioned, a lot of that could probably have been avoided.

Judging from Voxeltree's comments though, he is a new user (if not completely new). That being said, I am still hard pressed to find out why this would be important than choosing the right tool.

For example, when I worked on space applications and R&D, we preferred ABAQUS due to its nonlinear capabilities. When working large static models, the aircraft industry prefers NASTRAN.

The point being that different software packages often have a niche and if you are working in a specific industry or area within an industry the package itself is probably more important than whether it can multi-thread or not (btw, I am pretty sure its largely moot since all major ones are updated for this).



Brian

http://www.espcomposites.com

 

[ESPcomposites]

Voxeltree,

You appear to be new to the different options. There are also dedicated forums to NASTRAN, ABAQUS, etc. Many posters have been using them for years/decades. They can help with more specific questions. The ABAQUS forum appears to be more active than the NASTRAN one though.

Brian

http://www.espcomposites.com

 

[voxeltree]

http://www.youtube.com/watch?v=2qy5H0eZdJk

 

[voxeltree]

I'm not chooseing the software based solely on its multi-threaded capabilities, but that is a very important option. Thats why they usually charge extra for that capability. The goal of my search was to find a great FEA app to suit my purpose, which is large static models used in dragline construction. Sorry I was not more explicit.

 

[FeX32]

I meant to say Comsol !

It's a great package. Check it out. Get a copy of the trial.


Fe

 

[cbrn]

Hi,
all hi-end FEA packages are scalable on any number of processors you want, or so.
At least, ANSYS does. And it does in several different ways, so that you can even build computing arrays and solve in remote with dedicated hyper-parallel solvers.
It does charge you for all that, though.
If you want to have a look on Ansys, search for "HPC" (stands for "High-Performance Computing").
Regards!