[lokupznm] AES random number generator

With AES instructions built into recent x86 CPUs, is an AES-based pseudo-RNG competitive in speed against good versions of traditional non-cryptographic random number generators (e.g., linear congruential, lagged Fibonacci, Mersenne Twister)?

We can probably get away with reduced round AES for non-cryptographic applications. How many rounds do you need to avoid sampling problems in Monte Carlo simulations?