Any fastest subroutines for Parallel Eigenvalue computation?

[GoldenRetriever]

My program spends major of time on calculating the Eigenvalue soI'm looking for asubroutine for Parallel Eigenvalue computing in hope of shortening that part.

I could not find anywhere in the IMSL library which I currently have mentions about that so I assume it is not available there.

If anyone have tried any subroutines or have any experiences withParallel Eigenvalue sburoutine please advise.. Will appreciate anysuggestions. Cheers

[Steven_L_Intel1]

Have you looked at the Intel Math Kernel Library? This is included with your product and MKL has quite a few optimized Eigenvalue routines. Ask in the MKL forum if you have further questions.

[mecej4]

We need a more complete description of your requirements. There is no such thing as the eigenvalue -- a matrix of size n X n has n eigenvalues. If you want the eigenvalue of largest magnitude, that of least magnitude, etc., you need to specify that clearly.

If you have some information about the spectrum of the matrix and desire to compute many eigenvalues, you could consider making a number of copies of the matrix in order to facilitate computing multiple eigenvalues. If you are able to estimate some eigenvalues, you could consider an iterative refinement method for computing nearby eigenvalues.

[GoldenRetriever]

Thanks Steve andmecej4,

I would like to find the eigenvalues and eigenvectors of A*x=lamda*B*x problem. The matrices A, B are band maxtrices (Symmetric or Nonsymmetric) size nxn with n may be between 5000 to 9000.

Follow Steve's suggestion, I'm looking into the driver routinep?sygvx in MKL, hopefully I'm moving in the right direction? Cheers