Iterative Solvers in Industrial CFD codes on Modern Microprocessors.

Preconditioned conjugate gradient iterative solvers are widely used in today’s popular industrial computational fluid dynamics (CFD) codes. The most popular industrial codes use unstructured grids, and employ finite volume discretization to solve the transient 3D Navier-Stokes equations for fluid flow. Computationally, the simultaneous solution of the algebraic equations set is the most time consuming aspect, and iterative solvers are almost invariably used. The accuracy, speed, and robustness of the solvers have allowed CFD to become an integral part of the design and manufacturing process. The codes have also evolved into distributed memory parallel applications and no longer require any shared memory parallelization. These characteristics have several important implications for maximizing performance on modern computer microprocessors. These characteristics and their historical basis will be discussed in this paper.

Mr. Stephen R. Behling

IBM Corporation

392 Woodlawn Av.

St. Paul, MN 55105

612-626-8926 (wk)

651-698-3921 (hm)

sbehling@us.ibm.com

(Optional title: CFD Application Specialist)