Comparative Study of High Performance Computing Using Multi-core Parallel Systems

Multi-core based high performance computing systems are available with a reasonable price. Parallel programming paradigm needs to be adjusted to an individual system. Parallel computing systems were compared in this paper. Electroencephalography signals were collected in order to measure performance of parallel computing for CPU and GPU based systems. A CPU based system showed better performance for smaller data set, while a GPU system showed better performance for larger data set. GTX580 processor, which has 512 CUDA cores, showed consistent speedup as input data was increased continuously. However, CPU has a limited speedup due to the lack of parallelism. For the FIR filter computation, GPU showed a good scalability, while a CPU system did not. The performance of GPU was better than CPU system slightly.