Argonne National Laboratory

Accelerating Performance of NekCEM with MPI and CUDA

TitleAccelerating Performance of NekCEM with MPI and CUDA
Publication TypeConference Proceedings
Year of Publication2013
AuthorsMametjanov, A, Min, M, Norris, B, Hovland, PD
Conference NameSuper Computing (SC13)
Conference LocationDenver, CO
Other NumbersANL/MCS-P4064-0413
AbstractModern supercomputing architectures are increasing the number of cores per node by adding GPU coprocessors to increase instruction and memory throughput. Compute-intensive applications need to take advantage of higher throughput by employing both distributed and shared memory programming models. In this paper, we describe our approach for accelerating Computational ElectroMagnetics application NekCEM using MPI and CUDA programming models and demonstrate its performance on the leadership- class computing systems. The resulting code kernels can potentially be reused in other applications within the Nek family of codes.