The most essential for us are ZGETRF and ZGETRS and the MAGMA and ParLAPACK equivalents. Our codes are in FORTRAN.
We have versions which run to solve large problems on an array of computers using ParLAPACK and would like to be able to do the same thing on a series of computers each with one or more GPGPU. The problems get so large that the matrix is too big for one computer's memory.
One possible approach (which I think I mentioned before) would be to put ParLAPACK on top of a MAGMA Lapack.
P.S. I actually meant ScaLAPACK, not ParLAPACK. Sorry for the mistake.
Last edited by fletchjp
on Mon Sep 24, 2012 10:56 am, edited 1 time in total.