NX Nastran multi core hyperthreading

[tstanley]

Hi:

I have been looking at buying a new computer for Windows 10 and have looked at the 12 core AMD processor. On the other hand previous threads, thread 561-357474 and thread 561-314150 indicate that the 12 core CPU would be a waste of money. John Baker replied in those that the math was not amenable to hyper threading but if you had the NX Nastran Advanced that you would be able to take advantage of it. Blas Molero also gave a useful reply.

The release notes for NX Nastran 2019.2 also indicate the same thing.

Will this still be the case in the near future?

It is quite depressing to run a "Basic Non Linear" problem and have it take overnight or even a couple of days to solve with my old lap top computer with 2 cores and 4 threads. It is even more depressing to visit my local computer stores full of computers that can take advantage of more cores just to run computer games.

Tom Stanley

 

[tstanley]

Further to my question:

I ran an analysis a couple of weeks ago using SOL 106 with about 1.3 x 10^6 degrees of freedom and it took 16.5 hours. I have read that there is a command dmp=p to run "distributed memory parallel processing" for SOL 101, 103, 108 etc. but no SOL 106. However the published documents that I have found are several years old. I haven't tried the command on my existing computer with only 2 cores to see if it runs yet but I would hate to buy a CPU with fewer cores and use more ram, only to find out that next year's version will be able to access 12 cores or more. It is easier to add ram later than to upgrade the CPU. The other thing is the 8 core version has 40 Mb cache and the 12 core version has 70 Mb cache.

Or would the SMP (shared memory parallel) command be more applicable to the multi core processors. The quote that I read about that says that performance tapers off above 8 cores.

I agree that adding ram and using M.2 SSD drives give more bang for the buck, but it just seems that the productivity of the software could be improved by using today's CPUs as well, and those that are to be introduced soon.

Tom Stanley

 

[JohnRBaker]

If you don't get the help you need here, you also many want to consider posting your questions in the dedicated NASTRAN forum:

https://www.eng-tips.com/threadminder.cfm?pid=825

John R. Baker, P.E. (ret)
EX-Product 'Evangelist'
Irvine, CA
Siemens PLM:
UG/NX Museum:

The secret of life is not finding someone to live with
It's finding someone you can't live without

 

[Tingsryd]

Is Nastran floating point calculation intensive? If yes you may want to take a closer look at processor architecture. AMD's offerings a while back shared a FP processor between every two cores, and FP performance was underwhelming because of it but I am not familiar with their current offerings. Intel Xeons are known for good FP performance and until the Core i9 I don't believe any Intel Core processors could keep up to Xeons without extensive overclocking.

I use NX for cad but mostly cam and the docs I've read about HT all recommend disabling HT. Don't let that discourage you though, it's all up to how well NX Nastran performs in proctice. My colleagues use a competing cam software (Master something or other...) which utilizes HT but in practice it still takes them longer to process similar operations and quite often they have to process small parts of a model to reduce time rather than the entire model. Processing the entire model is normal for NX cam and it excels quite handily without HT optimization.

There is multi threading support in NX for many functions, especially in cam, and every release has better core support so I would have to think Nastran is moving in that direction as well. If you post in the Nastran forum perhaps a developer could chime in about the Nastran roadmap regarding multi core support.

I used to get NX cad cam roadmap information from my PLM World RUG events but now that's a thing of the past since Siemens killed off RUG events.


NX 12.0.2
EVP's

 

[tstanley]

Thanks to both of you!
I have looked at the general Nastran forum for previous questions. I did learn more about dmp and smp. It looks as though smp would be available to me. I don't know if it uses floating point calculations. I will look into that.

The work that I am doing involves receiving the results of physical testing (4 bar test)of composite panels that use "linear" foam that has a high elongation (up to 50%) before breaking. The stress strain curve is only linear for part of the foams stress strain curve. This foam is used often because of this property. I can take these results and try to replicate them using NX Nastran Basic Non Linear. The ultimate goal would be then to simulate an actual real world part (the cockpit of an F1 racing powerboat)under real world loads from accidents, but the time involved in solving that would be enormous.

I know that the Basic non linear uses an iterative process to solve so that it would not be able to solve the iterations in parallel, but I have been hoping to run each iteration faster. There can be quite a few iterations, so that if each iteration could be run faster that would help.

From looking at the diagrams showing the difference between dmp setups and smp setups, it seems as though the multi core, hyper threading CPUs match the smp setup rather than the dmp setup.

Thanks again for your help. I am retired like John Baker and do most of this work on a volunteer basis. It is interesting work, I will just have to find a paying customer to justify buying the Advanced version when my license comes up for renewal.

Tom Stanley