In an evaluation I am currently undertaking I am finding that the Intel
version of BLAS/LAPACK for Windows (MKL) computes eigenvalues &
eigenvectors of unsymmetric matrices (DGEEV) of O(2000) some 30% faster
than the reference implementation compiled with CVF.
In my case this represents a meaningful saving of some four to five
minutes of wall clock time per matrix.
David.
Erik Schnetter wrote:
>
> On Monday 09 September 2002 11:33, you wrote:
> > On machines without a vendor version (e.g. Linux):
>
> There are vendor versions for IA32 Linux. Intel does provide e.g. the
> Intel Math Kernel Library, while seems to contain BLAS and LAPACK
> routines; see http://developer.intel.com/software/products/perflib/.
>
> -erik
>
> --
> Erik Schnetter <[log in to unmask]>
> Web: http://www.tat.physik.uni-tuebingen.de/~schnette/
>
> My email is as private as my paper mail. I therefore support encrypting
> and signing email messages. Tell me if I can encrypt messages to you.
> (For more information, go e.g. to http://www.pgpi.org.)
|