Solidworks Simulation 2009 runs too slow - what could be the reason?

[sprashanth]

Our company recently purchased a couple of HP workstations with Quad-Core processor and 8GB RAM to run High-mesh FEA simulation on our CAD models, since we were unable to do so effectively with our existing dual-core systems with 2GB RAM. But unfortunately, the new workstations have not yielded significant improvement over the old ones.

For example, our CAD model upon preprocessing with High mesh has typically about 100,000 nodes, 500,000 elements and above 2,000,000 degrees of freedom. On our existing Dual-core systems, it would take about 4 days or more for the simulation to run & on the new workstations, it takes 3 days to run - which isn't that significantly better, considering we have double the processor speed and four times more RAM. Our requirement is to finish each FEA within a day, and 3 days for each is just too long.

What could be the reason for the shortcomings of the new workstations? Is there some software or harware settings we need to set so as to maximize the performance, or are we just expecting too much out of the new workstations? By the way, we have installed Windows XP 64-bit operating system on the workstations.

 

[GRF]

Not every part of the FEA calculations take advantage of multiple cores. If you watch the resource meter in Windows you will see quite often only 1 procesor will be at 100%. There is not much you can do for those situations.

 

[GregLocock]

More accurately, the solver itself is single threaded. It is possible to multithread an FE solver, but typically you will be paying for a license for each cpu that you use.

Even with the most optimistic assumptions about how computers work the new computer was not going to be 4 times faster than the old.

Also check whether you are using virtual memory - if you can move the entire solution into RAM you will see an enormous speed benefit - I think I saw a factor of 8 or more when I fitted enough memory to my laptop. Any sensible FEA system will tell you how much memory it needs for a given problem size.

And if I can point out the obvious - your models are enormous, do you really need that many elements, or are you just pressing automesh with a uniform mesh size? We used to build the bodyshell of a car with 70000 elements, that was fine for linear work.

Cheers

Greg Locock

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

 

[sprashanth]

@GRF:

I heard that meshing was supposed to be the single-core operation, whereas the solver would actually utilize all the processors during the operation.
Anyway, I will check out what the resource meter on the system shows while running an FEA and get back to you.

@GregLocock:

I would like to try out your suggestion of moving the entire solution to RAM memory, without using virtual memory. Can you explain how to do it? What settings do we need to change in order to do so? If would be great if there is a major improvement in the speed of the solver by utilizing wholly the RAM memory.

Regarding the model, we do automatic meshing using Solid Tetrahedral 2nd-order elements whith a uniform mesh size, but we do give mesh controls to the small and/or critical regions as necessary. We would like to go for Manual meshing, but Solidworks Simulation does not offer that option.

By the way, I would like to know how you could mesh the whole body of a car using just 70,000 elements - was it using solid or shell mesh, and which element type? Just curious to know.

 

[GregLocock]

Last first, it was a mixture of beam elements and plates. I didn't do it.

You can't force the model to use real RAM, you just have to reduce the size of the model until it fits the RAM you have, or increase the amount of RAM in the system.

Cheers

Greg Locock

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

 

[CorBlimeyLimey]

If the VM is being used and causing the slowness, you have some options;

1) Reduce VM to zero and add more RAM.
2) Install a separate SSD and use it solely for VM.
3) Install a separate fast HDD and use it solely for VM.

 

[clementfocuser]

I am in the same situation as last week I purchased Solidworks 2009 with nonlinear capability. I am now looking at the right kind of workstation to run nonlinear optimizations of my product. Dell precision workstation is what I am looking at. Is there any advantage to getting a quad processor? I am getting 64 bit XP but how much memory?

 

[dlafrance]

I have been working with a Dell Precision T3400 machine using XP64 for about 6 months and had nothing but trouble, I even descovered new bugs. At least, Solidworks has solve some of them but too few people work with XP64, this is why it is not stable. We will revert to XP32 to be able to work with less bugs. Solidworks 2009 is not stable on XP64.

Unless you frequently use more than 3.3 gigs of RAM, stay with XP32.

I can also confirm that simulation does not use the multi processors but photoworks does! Go figure!

 

[kellnerp]

The obvious question is why 2,000,000 nodes? I mean an analyst in this situation would figure out a way to get the dof count down.

You didn't mention whether this was non-linear or linear analysis. If it is non-linear you might consider getting a RAM hard drive because you will invariably choke on disk IO. Don't get a flash based RAM drive either. You need one that has fast RAM.



TOP
CSWP
BSSE

http://www.engtran.com

http://www.niswug.org

"Node news is good news."

 

[SalamiVice]

This is a little late, but it might help some.

I have come to the conclusion that SolidWorks simulation does not care on how much ram you use.

we run a lenovo workstation it has 16 gigs of ram, SSD hard drive, 8mb of L3 cache per core, and I have not seen any real benefits over the dual core we use for normal operation with 3.3 gigs of ram.


only solution it to reduce your nodes and DOF. SW2009 has a nice feature "Simplify Model for Meshing" use this where you do not care about results. this should enhance the speed of your simulation.

 

[kellnerp]

If you want to use multiple processors look at NEiWorks or NENastran. They can run not just multiple processors but on clusters. SRAC once told me they would be able to do this, but it sounds like it isn't happening.

This must be a non-linear analysis because normally 2million dof should solve rather quickly.

What is the bandwidth reported in your .out file?

TOP
CSWP
BSSE

http://www.engtran.com

http://www.niswug.org

"Node news is good news."