SOLUTIONS FOR IMPROVING THE PERFORMANCE OF RANDOM NUMBER GENERATORS USING GRAPHICS PROCESSING UNITS

In this paper, we have researched the possibility of using graphics processing units (GPUs) for generating a variety of random numbers. We have compared the performance obtained on the latest GP Us architectures with the one recorded on a high-end central processing unit, we have analyzed the main features of each architecture that influence the performance, we have developed solutions for optimizing the performance of random number generators. Although there is a great interest lately in implementing random number generators on parallel architectures, none of the works so far (to our best knowledge) has developed and studied implementations on Kepler, the latest Compute Unified Device Architecture (CUDA) architecture. We have developed a high performance implementation, harnessing the computational resources of the latest Fermi and Kepler architectures, studying their optimization potential, their impact on the execution time, on the generated samples bandwidth, on the energy consumption and on the execution cost.