Parallel Implementations of Multilevel Fast Multipole Algorithm on Graphical Processing Unit Cluster for Large-scale Electromagnetics Objects

This paper investigates solving large-scale electromagnetic scattering problems by using the Multilevel Fast Multipole Algorithm (MLFMA). A parallel implementation for MLFMA is performed on a 12-node Graphics Processing Unit (GPU) cluster that populates NVidia Tesla M2090 GPUs. The details of the implementations and the performance achievements in terms of accuracy, speed up, and scalability are shown and analyzed. The experimental results demonstrate that our MLFMA implementation on GPUs is much faster than (up to 37x) that of the CPU implementation.