5
$\begingroup$

Is it possible to use the Parallela computing platform to enhance Mathematica computing on a Raspberry Pi?

It's not at all clear to me how to interface the two boards (here it says it's done via Ethernet), but in this video it appears someone used Parallela to run videogames on a Pi. I doubt that the videogame was custom made but my ignorance in parallel computing is appalling.

Mathematica has built-in support for parallel computing, even on a Pi, even though it appears to refer to Pi clusters.

So I was wondering, has anyone tried the mix MMA+RasPi+Parallela? Does that even make sense?

$\endgroup$
  • 1
    $\begingroup$ The video shows the game running on the ARM processor only; when the Raspberry Pi is connected, it is the Parallella that uses the Raspberry Pi as an accelerator, not the other way around (as the Parallella has no OpenGL hardware). None of these examples show the multi-core "Epiphany" chip in use, which does seem to require code to be written (probably in C) to take advantage of it. Parallella is architecturally similar to Raspberry Pi and so can perhaps run Mathematica, but unless one writes one's own code as well this doesn't seem to provide much benefit. As a point of comparison, ... $\endgroup$ – Oleksandr R. Dec 5 '14 at 10:08
  • 1
    $\begingroup$ ... Parallella provides 20 GFLOPS in single precision, whereas a typical desktop computer can achieve almost 200 GFLOPS double precision, and GPU roughly similar (or about 1 TFLOPS single precision). So, with Parallella, you have something with about 1/10 to 1/50 the theoretical performance of standard hardware, and an architecture that will make it more difficult to achieve a large percentage of the theoretical maximum. It can surely be useful for something (especially if they release the 64 core, 100 GFLOPS chip), but I personally wouldn't rush to buy one. $\endgroup$ – Oleksandr R. Dec 5 '14 at 10:22
  • 3
    $\begingroup$ Finally, the Epiphany cores are not ARM and focus solely on floating point tasks, so cannot run Mathematica (which anyway requires double precision). At best, as I see it, someone may write a BLAS library that can run on Epiphany and then it would be theoretically possible to produce a single-precision build of Mathematica that could use it; otherwise, hand-coding of your target algorithm in C or even assembly language would be required. The Parallella platform does provide OpenCL support so perhaps this could be useful in a Mathematica context. $\endgroup$ – Oleksandr R. Dec 5 '14 at 10:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.