topic Are the Iterative Sparse Solvers parallelled? in Intel® oneAPI Math Kernel Library

Are the Iterative Sparse Solvers parallelled?

watercube — Wed, 19 Jan 2011 00:04:30 GMT

Hi all

I have a question about the Preconditioned Iterative Solver, such as combining ILUT and FGMRES for non-symmetric matrix.

Is it parallelled by OpenMP in the MKL version 10.3?

Such as the direct solver PARDISO, it is said "a speedup of up to seven using eight processors has been observed."

Thanks very much!

Kevin

Are the Iterative Sparse Solvers parallelled?

Gennady_F_Intel — Wed, 19 Jan 2011 04:06:03 GMT

Currently, the ILUT and FGMRES preconditioners are not multi-threaded into the current version (10.3) of MKL.There are Feature Requests submitted earlier. We are working on these issues.

--Gennady

Are the Iterative Sparse Solvers parallelled?

watercube — Wed, 19 Jan 2011 04:33:44 GMT

Gennady

Thanks a lot!

Right now my solver is too slow for non-linear equation, I think I have to develop my own parallel solver!

Kevin

Hi, I wonder if the sparse

yanpu_z_ — Mon, 29 Apr 2013 12:59:11 GMT

Hi, I wonder if the sparse iterative solver is parallelized in the lastest version of MKL 11.0? Is it still inder development? Thanks a lot!

Hi yanpu,

Alexander_K_Intel2 — Mon, 29 Apr 2013 14:41:41 GMT

Hi yanpu,

Currently MKL support only sequential version of FGMRES. But this solver implemented as RCI routines so multiplication on matrix A and preconditioner can be implemented via parallel version of routines - SparseBlas, Blas, Lapack or Pardiso functionality.

With best regards,

Alexander Kalinkin