Performance and scalability improvement achieved by GaspiLS in comparison to an original MPI implementation in an industrial CFD simulation. Shown is the scalability of the computation of a single PISO pressure correction step iteration on a grid with 1.1 million cells. The solver is a Jacobi preconditioned Conjugate Gradient method. In case of GaspiLS also the scalability on a bigger grid containing 7.8 million cells is shown. Architecture: Intel IvyBridge, two sockets per node, 10 cores per socket, Infiniband interconnect.
Please visit www.gaspils.de for more information