GPU/CUDA-Ready Parallel Monte Carlo Codes for Reactor Analysis and Other Applications
Tianyu Liu, Lin Su, Aiping Ding, Wei Ji, Christopher D. Carothers, Xie George Xu, Forrest B. Brown · 2012
Monte Carlo simulation is widely used in reactor analysis and medical physics for its high accuracy. The time-consuming process of achieving a desired precision can be significantly facilitated by parallel computing methodologies. The exascale High Performance Computing (HPC) architecture, which is expected to arrive within this decade has the potential to further accelerate the Monte Carlo calculations. However, the ever-increasing parallelism is facing at least two challenges. One is the fact that the power consumption increases with the Floating Point Operations Per Second (FLOPS). The other is a lack of software for the emerging hardware. The first issue is addressed by seeking more energy efficient system, such as Graphics Processing Unit (GPU)-accelerated heterogeneous systems. For example, Tianhe-1A and TSUBMAME 2.0, which---ranked as the 2 and 5 most powerful computers in the world by November 2011---consists of thousands of NVIDIA GPUs. This paper addresses the need to meet the second challenge, which is to develop a Monte Carlo code that can be effectively implemented on GPUs.