A Comparison of Graphics Processor Architectures for RFID Simulation

摘要

Graphics Processing Units (GPUs) have a huge number of cores to speed up graphical computations and they are being used in a wide area of general-purpose applications that require high performances. In this paper, GPU computing is exploited to model the signal propagation and the interference in large RFID systems, which are a promising solution for achieving pervasive computing since they offer the automatic object identification. The speedup of the parallel algorithm is evaluated with respect to a sequential version. Two popular frameworks for general-purpose computing on GPU are considered in the comparison, i.e. CUDA and OpenCL, and distinct implementations are provided for them, highlighting their differences in code optimization and performance.