Hey all, I want to purchase a workstation for my research work within a budget of $8-10k. My work is mostly heavy CFD stuff with some EM. I wanted to see the best configuration that would fit my need. I was inclining towards the threadripper pro 7975WX and some of the Xeon W7 processors with a 128 GB or more RAM, but the discussions on the other forums have confused me, mainly because they go over my head given my expertise in this domain xD.
It would be great if there are some recommendations that would work best for my use case. Any help is much appreciated!
Unfortunately Power9 offers pretty weak performance, be it due to unoptimized code or just an antiquated architecture, you can expect less than Sandy Bridge performance in many tasks on it.
Comsol actually use to be compiled for Itanium and PowerPC architectures.
I’m tempted to put up a “vintage computer” section of scores on that thread and benchmark a HP ZX6000 and Apple Power Mac G5 I have.
Thanks! I just saw it, I understand that it is something like an interplay between how much memory you have, how fast is your processor and how much memory you can allocate in the processors…not sure but I think its something in those lines.
But many of the options available to me for a purchase are not available on that list. I am specifically looking into straight up models that can be built on dell or lenovo towers…so they offer something like Xeon W5 or W7 or W9 or Threadripper 59XX 79XX or so on.
Is there a way to extrapolate/intrapolate the reults of the benchmarking given in that post to these processors?
PS: I might sound silly asking but trust me i dont know anything about computers
Out of all those choices you’d probably be best served by a Threadripper 79x5WX machine (or even better would be the newly released TR 99x5WX), it’ll do the best in a wide variety of problems only loosing out to the Xeon W7/W9 in some of the more dense matrix MEF problems. The Threadripper will have a particularly high advantage over the Xeon in pure-CFD problems.
One thing to look out for on both of these platforms/processors in memory cooling, I’m not sure if Dell or Lenovo offer a workstation with fans over the memory cooling it, but if they do that would help me choose one over the other.
I have a 2-socket Xeon gold 6136 system w/ 384G memory and a 32-core 7975WX w/ 2TB RAM. Running a model that fits on both machines (3D multiphysics w/ unidirectional coupling from CFD to thermo-mechanical), I find the AMD machine is consitently 3x faster. On small 1D models this drops to 2x faster, mostly because the solver setup time is dominated by single-threaded tasks. I think it could go even faster if I wasn’t hamstrung using JDEC timings… unfortunately my org forced me to buy a Dell instead of letting me spec my own Threadripper system, so I’m stuck with that…