Graphical Processing Units (GPU) acceleration of finite-difference frequency-domain (FDFD) technique

The evolution of the Graphics Processing Units (GPU) driven by the computer games business brought a graphics hardware as a high-performance, programmable and non-expensive chips. Nowadays, the graphic card has a truly programmable architecture which allows to process data with high parallelism and high memory access rate. That is the key motivation fact for using them for parallel processing. The performance of Graphical Processing Units (GPU) has been significantly improved in recent years. Compared with the CPU, the GPU is better suited for parallel processing and vector processing and has evolved to perform various types of computation, in addition to graphics processing, including numerical computations. In general, there are two factors that in some combination make GPU attractive target architecture for accelerating general purpose computations. First, is the raw throughput speed of GPU compared to CPU. Second, the GPU acts as a co-processor, therefore it can free up the CPU to perform other tasks.The graphics processing unit (GPU) has been used for the solution of electromagnetic scattering problem using the FDFD method. The radar cross section for different structures has been calculated using the FDFD code implemented using Matlab for the CPU and implemented using the BrookGPU platform for the GPU. The solution obtained by using the GPU code showed more than three times speed over the CPU code.