Monte Carlo methods rely on sequences of random numbers to obtain solutions to many problems in science and engineering. In this work, we evaluate the performance of different pseudo-random number generators (PRNGs) of the Curand library on a number of modern Nvidia GPU cards. As a numerical test, we generate pseudo-random number (PRN) sequences and obtain non-uniform distributions using the acceptance-rejection method. We consider GPU, CPU, and hybrid CPU/GPU implementations. For the GPU, we additionally consider two different implementations using the host and device application programming interfaces (API). We study how the performance depends on implementation parameters, including the number of threads per block and the number of blocks per streaming multiprocessor. To achieve the fastest performance, one has to minimize the time consumed by PRNG seed setup and state update. The duration of seed setup time increases with the number of threads, while PRNG state update decreases. Hence, the fastest performance is achieved by the optimal balance of these opposing effects.
This paper examines the performance of two popular GPU programming platforms, Numba and CuPy, for Monte Carlo radiation transport calculations. We conducted tests involving random number generation and one-dimensional Monte Carlo radiation transport in plane-parallel geometry on three GPU cards: NVIDIA Tesla A100, Tesla V100, and GeForce RTX3080. We compared Numba and CuPy to each other and our CUDA C implementation. The results show that CUDA C, as expected, has the fastest performance and highest energy efficiency, while Numba offers comparable performance when data movement is minimal. While CuPy offers ease of implementation, it performs slower for compute-heavy tasks.