I'm quite new to GPU or in general parallel programming. I am working on a Lattice-Boltzmann simulation that should be parallelized on GPU's. I wanted to asked more experienced people how to attack such a project. Would it be more beneficial to first implement all necessary features (basic algorithm with some extensions) and than to tackle the parallelization or should it be better done part by part. Speaking first the basic algorithm, than parallelizing this part. And then the same procedure for the extension?

Thank you in advance!

  • 2
    $\begingroup$ It is not useful to parallelize all parts of an algorithm. You first need to know where your compute time is spent, and then address that part by parallelization. $\endgroup$ Jan 4 at 18:21
  • $\begingroup$ At some point you'll have to decide what programming language to use for this. A higher level language like C/OpenMP or C/OpenACC will make implementing a solution much easier and make your code more portable than trying to write your code in a much lower level language like CUDA. $\endgroup$ Jan 4 at 22:19

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.