Any library can help me to solve the eigenvalue of a big sparse matrix with cuda
and MAGMA:
[url]http://on-demand.gputechconf.com/gtc/2014/presentations/S4541-magma-linear-algebra-gpus.pdf[/url]
are two GPU-accelerated libraries you may want to look at.
There is of course also cusparse, and SuiteSparse is starting to add GPU acceleration to some functions:
[url]http://www.cise.ufl.edu/research/sparse/SuiteSparse/[/url]
Thank you for help txbob.
The first lib CULA. Its sparse tool isn’t free probably. I can’t download it. MAGMA is great lib. But I can’t build on windows. That’t too bad.
I hope cusparse can solve in the future. Now I am trying MAGMA and slepc on linux.
Did you know any other lib can solve it on windows with cuda?
Any way, Thank you indeed.
How big is the sparse matrix and what is the number of nnz ? If it is small enough (or dense enough) could you use the dense libraries (like MAGMA, which the newest version does exist for Windows).
There is an eigenvalue example in the CUDA SDK, and you can look at that code for the general idea of how to implement. The main issue is to implement using the sparse data formats I would assume.
My matrix is 190000*190000. It is a small matrix in my study. At same time the matrix is sparse enough, a row may only have a few nnz. My study is about nanoscale with FDFD method.
Now I am working on linux with other lib. I hope cusparse can solve in the future.