More on solver speed: ROI of expensive commercial sw?

[mastroit2002]

Dear Sirs,

I'd like to ask You a new question on this topic which arises from the insight of given aswers.

I guess that my opinion about solver speed issue was not far from true: on today computer near all static/modal problems cam be solved quite easly in a proportional to precision ammount of time.

Only really tough problems (more expensive) like those due to non-linearity, speed time response (crash like) or CFD (high reynolds, transport phenomena, combustion ....) really need computing perormances.

I know that several commercial code offer several solution methods (explicit/implicit, direct/iterative), or even Multigrid/AMG methods.

Now here comes my question and iterest:
is there a Return On Investement of those codes one can be reasonably confident to achive?


I long for your prompt reply!
With best regards,

mastro

 

[GregLocock]

I disagree. There are many linear problems, such as fatigue life prediction for car bodies, that need a great deal of attention paid to solver speed.

I don't know about RoI, since even the most expensive code costs less than the analysts wages, and the cost of building the FEA model is less than the cost of getting the data to feed into it. Of course, if you run uncorrelated models then this will not be the case. But in that case you might as well use the back of an envelope.



Cheers

Greg Locock

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

 

[johnhors]

Greg is quite correct. The cost of software and the hardware used for FEA in industries I have worked in is insignificant compared to the value of the product being manufactured, salaries, material costs and so on. An enlightened management will know this and will purchase the best software that the job requires as well as upgrading hardware at least once a year, regardless of the price, all to remain as competitive as possible in the market place. It takes a short sighted and ill informed management not to see this, and ultimately such a management eventually destroys the company.

Believe me I have first hand experience of this happening!

 

[mastroit2002]

Many thanks for Your prompt reply!

I still have a question about the case study which really need speed perormances.

GregLocock reffers fatigue life prediction and to input data to feed FEM model, while johnhors claims about FEA cost to product cost ratio. I liked the insight and I got clearly the meaning of Your answers.

My question was more on the net weight of specialized codes on the total cost of product design: this was the sense of my question the return on investiment (commercial codes/ analysts wages) one can achive.

Even in easier way: which is the manifecturing area (defense, aerospace, medical, power supply) really earning from virtual prototyping with speacilezed codes?

Thank you twice!

Regards,

mastro

 

[rb1957]

for my money the key atributes of a FE package are going to be appropiate solutions (linear, non-linear, 2D, 3D, ...), accurate solutions (for my application), ease of user interface (both pre- and post-processing). these features can cost you alot of money very quickly. speed of analysis would be pretty lower on the list.

 

[GBor]

Solver speed would be lower on my list as well, but most of my models are not so complicated that speed is an issue. Some of the impact problems are involved, but I let them run over night if I need to. The time saved from an easy user interface more than compensates for any lack of speed.

Someone running "gazillions" of DOF's would likely have a different opinion.

 

[brad]

Automotive N&V and full-vehicle crash guys still put a premium on performance, as their models are often on the order of 2-10 Million dofs (and often thousands of modes in the case of N&V).

This is why at the automotive OEM's the niche players have almost no footprint--they can't scale to solve these sizes of models in a reasonable timeframe.

 

[GregLocock]

Another thing that bumps the solver requirements up is when we run DOEs, as they typically use a full factorial experiment, say, 7 variables at 3 levels.

Cheers

Greg Locock

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