PERFORMANCE APPRAISAL OF AN INTEL CORE i7 CPU AND AN NVIDIA QUADRO K420 GPU

Abstract

Nowadays, demand for High Performance Computing (HPC) is on the rise even though certain problem sets remain within the domains of Super High Performance Computing (SHPC) with the advent of applications such as Big Data analytics, weather forecasting, cognitive computing, quantum physics and climate research, etc. Hence, it behoves that sequential processing is undoubtedly no longer sufficient. NVIDIA, within the commercial realm of computation, has proposed a Compute Unified Device Architecture (CUDA) framework to harness the power of Graphics Processing Units (GPUs) for parallel computing which before now were only been utilized for Graphics Application. Recently, they are used for certain types of high performance computation. In this paper, a comparative analysis of sequential implementation on an intel core i7 Central Processing Unit (CPU) and parallel implementation on an NVIDIA Quadro K420 GPU for dense matrix-matrix multiplication was carried out. A look at the performances on different n square matrices using various grids of variable number of threads per block showed the GPU with its multiple cores provided significant reductions compared to the CPU with speedups of 114, 115, 270 and 941 for 128, 256, 512 and 1024 square matrices respectively. The authors recommended that the use of GPUs for computations with data reuse void of data dependencies should be exploited by scientists and engineers in the building of applications and systems.