I'm interested in buying a new workstation for my lab and in the process I encountered a lot of options which some I did not fully understand.

The main function this workstation is for numerical simulations which mainly involve: solving large eigenvalue problems and algebraic equations, large lists manipulations and setting (symbolic and numeric) and analyzing large databases.

I am not sure which hardware is needed for each usage (more memory, more CPU speed, more processing cores. etc.), so I can't really tell which system I need.

Furthermore, I am using parallel-computations in my simulations which cuts of a great deal of time for it to run. Thus, I was offered two alternatives: using coprocessors (such as Intel Xeon Phi) or using GPU processing (nVidia Tesla or Titan). I would like the advantages and disadvantages of each hardware while using Mathematica for my purposes (including the price issues). Moreover, I know that for complicated calculations CUDA programming is necessary for exploiting the most of GPU processing, are there any adjustments I need to make in order to use coprocessors with Mathematica or does it automatically using all available processing-cores?

After my little internet searching I was thinking to get something like the following:

  • Intel Xeon 8 core 2.4GHz processors X2
  • 128Gb RAM memory
  • Intel Xeon Phi coprocessor 51XX or 71XX (depending on price) X2/X3
  • OS SSD drive
  • 2Tb data drive X2

Is it too much? too little?

Thank you all in advance


closed as primarily opinion-based by m_goldberg, Bob Hanlon, gwr, Young, MarcoB Aug 19 '16 at 4:04

Many good questions generate some degree of opinion based on expert experience, but answers to this question will tend to be almost entirely based on opinions, rather than facts, references, or specific expertise. If this question can be reworded to fit the rules in the help center, please edit the question.

  • 1
    $\begingroup$ Depends really, can't you run a few examples and see with MaxMemoryUsed[] how much you are using? I personally never go beyond 16GB, with what I consider large datasets, but your definition could be different. $\endgroup$ – Feyre Aug 16 '16 at 11:00
  • 1
  • 1
    $\begingroup$ The way you access nvidia graphic cards (probably CUDALink) is different from how you access a Xeon Phi (OpenMP?) and we don't know if you are an experienced CUDA programmer or much more productive and experienced with the software tools available for Intel Xeon Phi processors. $\endgroup$ – Karsten 7. Aug 16 '16 at 11:47
  • 1
    $\begingroup$ I think you really should start from a consideration of what you currently do with Mathematica and determine your need from there; else you will end up spending a ton of money on resources you are hardly ever going to use. $\endgroup$ – Sascha Aug 16 '16 at 13:25
  • 1
    $\begingroup$ @Karsten7. I contacted them, will update once they give an answer. $\endgroup$ – Getz Aug 16 '16 at 20:15