Multigroup Monte Carlo on GPUs: Comparison of history- and event-based algorithms

•Improvements offered for both history-based and event-based MC transport on GPUs.•The history-based algorithm significantly outperforms the event-based algorithm.•Filtering dead particles to reduce thread divergence leads to 2x performance gain.•Performance on an NVIDIA P100 GPU equivalent to 60–80 CPU cores is demonstrated.•Nearly perfect inter-node scaling is demonstrated on the Titan supercomputer.