high performance computing on graphics processing units: hgpu.org

The Parallel finite difference time domain (FDTD) algorithm is an important method to 1 enhance the speed in multiple data FDTD operation. The improvement of graphics processing unit (GPU) performance, especially the emergence of Computer Unit Device Architecture (CUDA), offers parallel FDTD method an efficient and simple solution. First of all, this paper explains parallel FDTD method and CUDA in detail, and then elaborates the parallel speedup principle of two-dimensional FDTD algorithm by use of GPU on the CUDA platform. At last, this paper analyzes computing time of the FDTD algorithm on two different CPUs and GPUs. The result shows the consistency of the results between serial and parallel algorithms, and observably speedup is obtained compared with traditional PC computation by use of GPU.