2
$\begingroup$

I have a calculation which runs out of memory on my machine (Windows, 8 GB). I have already spent enough time on optimizing my code, now it's time to get some more power. I have read a lot of specific questions and resources, some fairly out of date, about the details of setting up Mathematica on cloud instances etc. Unfortunately none of those was complete or seemed simple enough for me to do in a short amount of time. On the contrary, I am under the impression that it is something which requires a lot of work and troubleshooting and that just isn't possible for me to do at the moment. On top of that, I know that I will not need it repeatedly in future, it is part of a one-time research and will require few hours in total.

The most interesting offer I found was this: https://www.nimbisservices.com/shop/product/mathematica-on-amazon-cloud-licensing-service/. I had some questions to the offer which I have asked and might post it here afterwards for the record. Other idea was to just get an improvement on my own hardware or link together several smaller machines myself.

What is the easiest way to get more computing power for a one-time computation?

$\endgroup$
  • $\begingroup$ This question is hard to answer without knowing anything about the code you want to execute. $\endgroup$ – Henrik Schumacher Mar 30 '18 at 14:30
  • $\begingroup$ @Henrik it's a Solve of a system of nonlinear equations and inequalities, saves the output to a file and that's it. $\endgroup$ – František Kaláb Mar 31 '18 at 15:49
  • $\begingroup$ @Bill is this really easier then getting a remote kernel? It makes sense somehow just makes me feel like I don't really want to have 64 GB of RAMs lying around $\endgroup$ – František Kaláb Mar 31 '18 at 15:51
  • 1
    $\begingroup$ How large is the system? Symbolic or numeric? How dense is the linearized system? Is there any structure that can be utilized to parallelize the task (with a reasonable amount of effort)? If not then getting access to a single CPU with lots of memory as suggested by Bill might be the simplest option. $\endgroup$ – Henrik Schumacher Mar 31 '18 at 15:54
  • $\begingroup$ @HenrikSchumacher, not that large, it's a computation on a 3x3 graph but with a lot of parameters, symbolic is a must, it's not sparse. I have already reduced it to a loop with a lot of smaller calculations, but still some of those calculations run out of memory. Now the question was already about the access, how should I do that the best, buy a lot of RAM in a store? Feels old fashioned. $\endgroup$ – František Kaláb Apr 1 '18 at 10:10

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.