Mulitple Cores and FEA vendors - licensing

[Flattie]

Hello, we use ANSYS and CFX at my work. We have new quad-core computers (64 bit 8 gigs). ANSYS requires additional HPC licenses to utilize the extra cores we have on our machines. ANSYS will use 2 by default (CFX only 1!), before requiring additional cash outlay for expensive HPC licenses that will allow all cores to be used. I am curious if other mainstream FEA providers use a similar multi-core licensing strategy? I have made it clear to our support provider that this practice seems like a money grab from ANSYS. After spending over $100k on the package only to find out our computers are 1/2 crippled without paying even more (this was never mentioned up front), leaves a bad taste to say the least. Anybody else want to chime in on this practice and whether you support it?

 

[jakeline6]

I have heard of this before with another company called CSAR which did it with their version of Nastran and I specifically stayed away from them because of it. I use NEi Nastran and they do not have this practice. It seems very unfair because if I purchase a faster computer should I also have to pay more? Ridiculous.

There are a lot of subtleties in licensing that people should be aware of such as charge for extra CPUs, ability to run the solver in batch mode outside the modeler, the ability to run multiple instances of the modeler or solver on the same computer when using a dongle, the cost of running on Linux versus Windows, just to name a few. Also, before you purchase find out what upgrade costs are going to be and get these in writing so that if you buy a limited version they do not gouge you on the upgrade down the road to the full system you really want.

 

[johnhors]

Perhaps you should investigate whether a free-ware package like CalculiX

http://www.calculix.de

( or on windows from

http://www.bconverged.com

) would meet your FEA requirements. You may be pleasantly surprised by what is available.

 

[GBor]

AMPS (

http://www.ampstech.com)

and Algor (

http://www.algor.com)

do not charge for this that I am aware of. If you go with Calculix (which is a really impressive freeware program), you may want to look at Roshaz (

http://www.roshaz.com)

for the reasonably priced pre- and post-processor...I haven't tried Calculix's pre-processor in a long time, but it used to be (4+ years ago) extremely user UNfriendly! NEiNastran is a very good product (

http://www.nenastran.com).

Ansys is expensive enough for the capability...surprised to 'hear' they charge for each processor [surprised]

 

[johnhors]

To run Abaqus using more than one processor also requires extra "tokens" to be available, which you have to purchase in advance for of course!

 

[Stringmaker]

Yes, I was not happy about this change at first either. We have a few distributed licenses at my company so it isn't a huge deal. However, were I the one paying for the licenses, in no way would I buy the HPC option.

Looking at it from the flip side though, you can now use Distrubuted Ansys w/ 2 processors free of charge. This gives you access to the AMG and DSPARSE solvers which are superior in performance compared to their SMP equivalents.

Being that you're running an Intel machine, see if you have an option in your machine BIOS which allows you to confugure the computer such that the operating system sees each physical CPU as a processor. Rather than each core as a CPU. You'll be able to tell easily by looking at Task Manager afterwards.

 

[Jordonlaw]

MSC Marc is the same; you have to pay to use more than ONE CPU ! This then becomes an expensive solution although there is a token system as well.

To be fair when I restarted using FEA in 2000, PCs were only just becoming quick enough to take advantage of practical FEA codes and were expensive for memory. Now I can get a quad core machine with lots of RAM for the same price as my single CPU workstation. Mine you, the quad core architecture is still not as efficient as the dual core so most of the extra capability is most in poor bus speed at the moment acoording to the benchmarks I've done.

 

[jbaylor]

This thread is timely. There was some discussion last week about running CalculiX in parallel on its forum. One of the solvers (SPOOLES) has a good implementation. It seems that some users are running a multi threaded version on a regular basis.

 

[Flattie]

Thanks guys. Keep it coming. It definitely seems like an industry-wide trend. I just wonder how this strategy can survive say five years from now when people have 20 core (or more) computers. So will that 20 core computer require 19 HPC licenses at $2500 each just to utilize all the computer functionality? It seems ridiculous, but that's where things are headed.

 

[jbaylor]

With MPI, I am not sure there is a technical limit to how many processors you can run a job on. The limits are the FEA methods and software. I have run explicit FEA jobs across quite a few processors with decent efficiency. Implicit solvers do not have near the scalability though.

 

[GBor]

I suspect the industry will be self-regulating. As Ansys charges so much, the industry will migrate toward these other packages whose performance is comparable, but whose cost is significantly less. What will probably happen in the long run is that the price of the average software package will go up a reasonable amount and everyone will run on multiple processors.

 

[Stringmaker]

I'm going to play devil's advocate for a bit. It seems that now nearly all mainstream commercial FE codes in one form or another hit you for using multiple CPU's. I I suppose I can understand this as I doubt software developers with the skills needed to make good parallel codes come at a cheap price. On the other hand, those that don't charge $$ for additional CPU's to be used probably do very little development to improve parallel performance. So either way you slice it I think it's a lose-lose situation...or you get what you pay for.

While scalability is limited typically after 2 CPU's...there is a huge benefit between 2 vs. 4. Hence, if you can justify it, I would ask my company to purchase one or two HPC licenses. A couple of hours saved time here and there certainly adds up in a hurry! You should have little problem quantifying it to your boss so long as it's practical.

Looking at things from a more grounded point of view, let's face it, most of us aren't solving huge models with a huge number of time steps all day, every day. I really don't see it as a huge issue...atleast not for me. I've just learned to make the most of what I do have available. My peve is IT's lack of support for 64-bit OS's on workstations and how old our "new" computers are! Why do large companys operate on a 2-3 year technology delay again? *end rant*

 

[GBor]

It seems that now nearly all mainstream commercial FE codes in one form or another hit you for using multiple CPU's.

Guess it depends on your definition of mainstream, but I don't think this is the case for implicit solvers.

On the other hand, those that don't charge $$ for additional CPU's to be used probably do very little development to improve parallel performance

I know at least two mid- to large-sized companies that would greatly disagree with you.

...let's face it, most of us aren't solving huge models with a huge number of time steps all day, every day. I really don't see it as a huge issue...atleast not for me.

Sounds like you've made some very valid points for those of you not solving huge models, but then, why are people that are doing linear static analyses even spending money on "mainstream" software?

If you are "playing devil's advocate", I thought I would fight back. Personally, I manage people that do analysis, so the question is more of a $/hour question for me...