GPU-accelerated 3D avalanche and streamer PIC-MCC simulations using pre-exascale supercomputers

The multiscale nature of lightning presents a number of computational challenges for simulation. Codes utilising a particle representation have been noted to provide high precision, but suffer long runtimes due to large particle populations. We present a massively parallel GPU implementation of a 3D PIC-MCC code for streamer simulations for community use, with higher particle capacities than seen in the literature. Implementation details are presented, along with scaling tests, physics benchmarking, and streamer simulation results. We demonstrate that the presented software scales efficiently to 256 GPUs, simulates accurate physics, and obtain streamer simulations without the use of typical particle-merging schemes, providing results in unprecedentedly high detail.