Monte Carlo Neutron Transport on GPUs Available to Purchase

GPUs have gradually increased in computational power from the small, job-specific boards of the early 90s to the programmable powerhouses of today. Compared to CPUs, they have a higher aggregate memory bandwidth, much higher floating-point operations per second (FLOPS), and lower energy consumption per FLOP. Because one of the main obstacles in exascale computing is power consumption, many new supercomputing platforms are gaining much of their computational capacity by incorporating GPUs into their compute nodes. Since CPU optimized parallel algorithms are not directly portable to GPU architectures (or at least without losing substantial performance gain), transport codes need to be rewritten in order to execute efficiently on GPUs. Unless this is done, we cannot take full advantage of these new supercomputers for reactor simulations. In this work, we attempt to efficiently map the Monte Carlo transport algorithm on the GPU while preserving its benefits, namely, very few physical and geometrical simplifications. Regularizing memory access and introducing parallel-efficient search and sorting algorithms are the main factors in completing the task.