Pseudorandom numbers are used massively in Monte Carlo methods, e.g. in physics simulations. The fastest random number generators, such as the xoroshiro or xorshiro families by Vigna (see eg here) are extremely fast, taking approximately three CPU cycles for generating a 64-bit random integer. Using parallel instances and the SIMD registers present in modern CPUs (SSE, AVX) the speed can be further increased by a factor 2-3.
I would like to know if there are any real-world applications for which random number generation is a speed bottleneck (when using a fast generator). It seems to me that modern generators are so fast that almost any processing of the generated random data is likely to take longer than the generation step itself.
