Efficient Usage of CPU and GPU Hardware Resources


In order to show the efficient usage of hardware resources (CPU and GPU) during electromagnetic (EM) simulations in WIPL-D software, the overall simulation process should be explained. WIPL-D electromagnetic simulation contains two main phases: the matrix fill-in which is followed with the matrix inversion.

For electrically small problems, where the MoM matrix has up to a few thousands of unknown coefficients (unknowns), the fill-in time is dominant. For electrically large problems, above 50,000 of unknowns, the matrix inversion time requires 80-90% of the total simulation time. The most important aspect of the matrix fill-in is to efficiently exploit the advantage of modern CPU hardware, which all have multi-core capabilities. In addition, WIPL-D GPU Solver is an add-on tool that efficiently exploits high computation power of nVIDIA CUDA™-enabled GPUs to significantly decrease EM simulation time. The acceleration is up to an order of magnitude for the matrix inversion phase.

As an example, the total simulation time is decreased (by using multi-core CPU fill-in and multi-GPU matrix inversion) to less than one hour for a problem with 200,000 unknowns.

Section: Miscellaneous

For full version of the document, please check the following pdf.