A highly-effective parallelization of statistical time-consuming tests of Pseudorandom Number Generators using CUDA

This paper focuses on parallelizing the most time-consuming statistical tests of the pseudorandom number generators for execution on the Graphics Processing Unit using NVIDIA Compute Unified Device Architecture. We propose new efficient parallel strategies for several tests that exhaust most time and hardware resources from a Statistical Test Suite for Random and Pseudorandom Number Generators of the National Institute of Standards and Technology. We show that these tests can benefit from the GPU solutions, leading to substantial improvements in speed-up even though keeping the accuracy of the test results. Our results reveal that the new parallel methods execute up to 200x faster compared to their sequential counterparts of the NIST.