While Computational Science did address a highly upvoted question about which types of computation lend themselves to using GPU resources, I'm reluctant to push the boundary into specific hardware recommendations (and software to support it).

At least without further encouragement!

If not here, then there seem to be a few other sites with even a worse fit. Of course StackOverflow takes on a wide range of topics (often not without bitter objections), and there is the Software Recommendation site. I'll throw out Software Engineering as one I'm also familiar with, and add that it doesn't seem a good fit.

I have a fairly specific kind of application in mind, integer arithmetic of moderately large size (thousands of digits perhaps, not millions). I mention this because GPU integer operations are natively of fairly small size (think vectors of "single precision"), and library functions undertake a lot of mapping necessary to make those into support for extended precision integers.

  • 2
    $\begingroup$ I too am of the opinion that hardware recommendations are off-topic on this site. I find it hard to imagine questions that lend themselves easily to a canonical answer for hardware, since the technology changes so fast that any answer would be out of date 6 months to a year later. $\endgroup$
    – Paul
    Nov 4, 2019 at 2:57
  • 2
    $\begingroup$ I don't think that the fast pace for technology changes make this kind of questions off-topic. $\endgroup$
    – nicoguaro Mod
    Nov 7, 2019 at 22:05

1 Answer 1


I would say, only a very small subset of hardware recommendation questions are on-topic at Computational Science SE.

I would advocate for the analogy with the "software packages" topic. According to our on-topic page:

Questions about software packages or languages used broadly in computational science (e.g., PETSc, MATLAB, Trilinos, LAPACK, SLEPc, R, NumPy, SciPy, Julia, Maple, Octave) except Mathematica (which has its own site now). In general, high-level questions (e.g., about language/package features) are best. Questions that are essentially about debugging a code sample, or about low-level language syntax are a poor fit for this site, and are usually closed; such questions should be asked on language-/package-specific forums.

Thus, if I were to formulate a paragraph for this page, I would write something like that:

Hardware Recommendations, ask on Hardware Recommendations unless it is very specific and focused on the functionality that directly effects the work of a particular scientific computing algorithm. Questions of the type "What is the best GPU for scientific computing?" or "What CPU I need to succeed for a numerical methods course?" are not good examples.

I am open to specific and tailored hardware recommendation questions; however, I am not sure that we will have a lot of them coming. I am not sure if our community can answer those questions better than Hardware Recommendations; however, they have only a few of those.

I certainly consider that these questions are off-topic on Computational Science and rightly belong to Hardware Recommendations:

and I was not able to quickly find any questions there that I consider being on-topic on Computational Science.

However, the question draft that you described seems interesting to me.

We had at least Research Computing proposal that never took off but would be certainly a better fit.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .