GPU acceleration of electromagnetic scattering analysis using the method of moments

In this paper, the GPU acceleration of the Method of Moments (MOM) solution process is considered. Scattering analysis using Rao-Wilton-Glisson (RWG) basis functions [1] is accelerated by making use of the Compute Unified Device Architecture (CUDA) by NVIDIA. The implementation discussed extends the work presented in [2] to overcome the limitations that the amount of device memory places on matrix assembly phase of the MOM solution process. The other phases of the MOM solution process are also considered, and include the GPU-based LU decomposition, which is also not limited by the amount of device memory, as discussed in [3]. This CUDA implementation allows for single precision speedups of around 300×, 5×, and 8×, for the matrix assembly phase, the LU decomposition, and total solution time, respectively, when compared to a multi-core CPU implementation. In double precision, the respective speedups are around 50×, 3×, and 3×. Furthermore, these speedups are maintained even for cases where the memory required for the impedance matrix exceeds the amount available on the GPU.