speed comparison of stl random number generators

In this post we will compare the speed of the different random number engines of the c++ standard template library. All random numbers in this benchmark are created with std::uniform_int_distribution, but with different random number generators. The quality of the random numbers themselves is not considered in this post.

The code for the benchmark can be found here: random_generators_speed.cpp.

I used gcc version 7.5.0 (Ubuntu 7.5.0-3ubuntu1~18.04) and an Intel(R) Core(TM) i5-2540M CPU @ 2.60GHz to perform the benchmarks.

Measurements

$ g++ -O3 random_generators_speed.cpp && ./a.out
minstd_rand   12.8 ns per random number
minstd_rand0  12.8 ns per random number
mt19937       17.3 ns per random number
mt19937_64    8.24 ns per random number
ranlux24_base 17.6 ns per random number
ranlux48_base 16.6 ns per random number
ranlux24      72.6 ns per random number
ranlux48      196  ns per random number
knuth_b       18.7 ns per random number

If only speed is important (and not quality of the random numbers), you should checkout xorshf96 or fastrand.