Institute of Electrical & Electronic Engineers
The authors have showed how to implement fast parallel LCG, LFG and WHG pseudorandom number generators using the divide-and-conquer approach on contemporary multi-GPU systems. In this paper they show how to design and implement fast parallel algorithms for linear congruential, lagged fibonacci and wichmann-hill pseudorandom number generators. The new algorithms employ the divide-and-conquer approach for solving linear recurrence systems. They are implemented on multi GPU-accelerated systems using CUDA. Numerical experiments performed on a computer system with two fermi GPU cards show that their software achieves good performance in comparison to the widely used NVIDIA CURAND library.