2
$\begingroup$

Does anyone know a good library that implements basic sparse matrix operations such as transpose, SpMV eigenvalues, etc. in GPU (cuda/OpenCL)?

$\endgroup$
2
  • $\begingroup$ Well, there's Cusp, and for SpMV, there's this technical report by NVIDIA, but I'm skeptical of doing sparse linear algebra on GPUs (see On the limits of GPU acceleration by Vuduc, et al, whose work is reputable). You're probably better off trying to do the sparse linear algebra on CPUs. $\endgroup$ Nov 30, 2012 at 6:33
  • 2
    $\begingroup$ Hi zimbra, welcome to scicomp! Can you give a few more details about your application domain, expected problem sizes, and whether you would like to take advantages of multiple GPUs on distributed nodes (like a GPU-based cluster or supercomputer)? $\endgroup$ Nov 30, 2012 at 22:31

3 Answers 3

3
$\begingroup$

Check ViennaCL. I use that library in many projects to improve performance of sparse operations and I am very pleased with the results. Remember that using GPU makes sense only for relatively large size of the jobs. Otherwise, it is not cost-effective.

$\endgroup$
0
3
$\begingroup$

Look at cupy. It supports basically all common sparse formats along with the common operations.

$\endgroup$
1
$\begingroup$

The cuSPARSE library is the official library developed by Nvidia and comes with the CUDA SDK, it implements a lot of sparse matrix operations.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.