1 Reply Latest reply on Sep 30, 2013 11:32 AM by Jared Conway

    Flow Simulation 2013 - CPU and distributed solving arrays

    Paul Beierle

      Hi all,

       

      I have been courting the idea of making a central 'supercomputer' to solve our ( a company I consult for) flow simulation studies in a more timely manner. They now have 5 Flow sim licenses so I'm thinking that I could build a discrete ripper system to force data from them cells with extreme predjudice.

       

      I've been building my own systems (and others') for over 20 years now so I'm covered on the logistics. However, I also realize that CFD is inherently a linear process. I've seen the nifty chart that Anna Wood posted last year regarding multi-cpus and am wondering if there is one of those for this years software.

       

      Essentially, I want to set up a rack box with 5 dual-cpu xeon (liquid cooled and overclocked to the apogee of stablity of course) MBs, each with a 64 or 128GB SSD and windows 8 (laugh if you must, it's grown on me). I'm running Server 2008R2 on the 'mothership' so I can rock Hyper-V and other nifty tools as well.

       

       

       

      Sound sensible?

       

       

      oh, Is it safe to say that pure clock speed is really the dominant factor in solving? Assuming modern bus and memory of course.

       

       

      Thanks

        • Re: Flow Simulation 2013 - CPU and distributed solving arrays
          Jared Conway

          i would agree that using a specific computer for solving is the way to go over using the network solver.

           

          ram, hdd and CPU are all drivers for solve time. ssd is a given. ram you should have enough so your whole process is in ram. CPU, check the solidworks KB for the chart you're looking for regarding solve time vs CPU. I think with the config that you've selected, you'll have a few CPUs taht won't be used to the level that you'll want them to be used for the dollars that you'll invest. 2014 has some additional functionality for multicore but i haven't seen any perf numbers yet. overclocking those CPUs further vs adding more would be better.

           

          the only thing i'm not sure about is making the solving computer as virtual machine. it would be better to be physical. but i might be misinterpreting that comment.

           

          something to also consider, with the high cell count, you'll ahve a lot of data to display on the "remote" terminal. make sure that computer has enough power to be able to do that. cpu, hdd, ram and gfx will all come into play. and remember, there is no replacement for setting up the problem and simplifying appropriately. throwing all the hardware in thre world at a problem that isn't, is just asking for frustration.