Application of GPU computing to the Characteristic Basis Function Method

This paper presents results of the process of adapting an existing software algorithm, which performs the Characteristic Basis Function Method (CBFM), to run on a Graphics Processing Unit (GPU). The CBFM is a highly parallel process, which lends itself naturally to exploitation of the parallel resources of the GPU. The initial results show great promise, with speed-ups above 90. These results can be expected to become even better with careful hand-tuning and optimization, and open countless possibilities: from increasing the size and accuracy of the electromagnetic analysis, to performing it on a conventional workstation.