14 Replies Latest reply on Nov 22, 2014 6:56 PM by Scott Kramer

    What hardware spec is the bottleneck in a non-linear structural simulation?

    Nathaniel Van Vliet

      Trying to see what I can improve to speed up my non-linear analyses.  I'm running an i7 3930k w/ six cores that it only seems to partially utilize.  RAM speed/latency?  SSD I/O?  Or something else?  Where is the hiccup?

        • Re: What hardware spec is the bottleneck in a non-linear structural simulation?
          Jerry Steiger

          Nathaniel,

           

          I'm no expert on SolidWorks Simulation or on the hardware, so take any advice from me with a whole shaker of salt. It depends on the simulation you are running and the system you are running it on. If you have sufficient RAM, so that the system doesn't have to swap in and out with the SSD, then it is probably your CPU speed that is the weak link. If you don't have enough RAM, then adding RAM would be the right thing to do. (I suspect that you want all of your RAM to be the fastest that will work with your CPU.) If you can't add enough RAM, then SSD I/O is going to be important.

           

          You may also find that your choice of solver is important. Some solvers work better on certain types of problems and some solvers use multiple cores more effectively. It could be that the slow solver that uses the most cores is fastest, or it might turn out that the fast solver on less cores is best. Or you may have no choice.

           

          Jerry S.

          • Re: What hardware spec is the bottleneck in a non-linear structural simulation?
            Greg Fowler

            I'm no hardware expert myself, but I was told that unlike regular modeling tasks in SW, simulation (and rendering, etc...) can make use use of parallel processing.  Thus a processor  a better parallel pipeline is better.

             

            In the same vein, you can set up distributed computing where a whole network of linked PC's crunch through the same problem in parallel.  Solving all those systems of linear equations with large matrices is very conducive to parallel computing.

              • Re: What hardware spec is the bottleneck in a non-linear structural simulation?
                Jared Conway

                parallel pipeline > there are several features and solvers that use this technology but not all

                distributed computing > this is not currently available in solidworks simulation

                 

                OP, this is very likely the reason for your solve times. in the end you just have to wait. this assumes that you have completely optimized your analysis setup and that you have not reached the max on your RAM (and the you aren't in page file) and that you have not maxed out the IO on your drives (if you are using ssd you are probably fine). I have also assumed that you have turned off hyperthreading.

                 

                check the solidworks kb for some more info on hardware for solidworks but our experience is that it is better to look at the way the problem has been setup than to throw more hardware at it.

              • Re: What hardware spec is the bottleneck in a non-linear structural simulation?
                Scott Kramer

                Flow Simulation is the only product that will utilize more than one processor core.

                 

                That being said, you should maximize processor speed, ram quantity and speed, and hard drive speed.  Processor speed is just a numbers game, get the fastest you can afford.  I've never seen more than 16GB used during any simulation, so I would recommend 24GB to keep away from any max ram thresholds, and again, the fastest you can afford.  For hard drive, you will benefit the most from a solid state drive.  The motherboard is also a key consideration because it's capabilities define the maximum processor and ram speeds and quantities.

                 

                Do yourself a favor, before investing a bunch of cash, look at your simulation setup and figure out a way to simplify it.  You need to get the element and node counts as low as possible.  Use symmetry, varying mesh sizes, eliminate contact pairs, and anything else you can think of to make it simpler.  Contact and non-linear situations will always increase processing time.