high performance computing on graphics processing units: hgpu.org

The technology of computational devices has been developed over several decades especially graphic processors which not only deal with graphic works but also compute scientific problems. This processor is suitable for parallel computations instead of using expensive high-end devices. Many research groups have implemented parallel computations using the MPI method with multi CPUs to solve difficult and complex problems. Lately, the CPU clock speed, due to technological limitations. Thus, the MPI method could not be speed up by adding CPUs. The GPU consisting of many threads is one solution for the technological limitation and can make the computation faster. Recently, a CUDA development environment has been developed by NVIDIA. It makes parallel computations easier for researchers dealing with complex problems. The FDTD simulation, a useful computational electromagnetic method was chosen for the parallel computation. In this paper the comparison between the MPI method and the CUDA method is discussed, then the FDTD simulation is tested using the MPI method and the CUDA method. Additionally, a simulation of particle movement in the FDTD simulation is implemented. This experiment enables the PIC simulation which needs a great deal of time due to the large number of particles. Since the GPU has a lot of threads to deal with these particles, it is possible to downsize the simulation time using the CUDA method.