Abstract. In order to perform high-performance Monte Carlo simulations of fracture in certain composite materials, we needed fast methods for generating deterministic random numbers. We made several design choices, and due to the fact that the entire simulation was to be done on both CPUs and GPUs, we designed new methods for fast implementation of the AES in the ECB mode on such architectures. This paper describes our algorithms and summarizes the performance results. In our implementation we were able to produce a speed of 78.6 Gbits per second on the GeForce GTX 480, which was 31-62% faster than the fastest implementations reported in the recent literature on similar devices.
Rone Kwei Lim, Linda R. Petzold, Çetin Kaya