Advancements in GPU-accelerated particle-in-cell simulations with iPIC3D

Particle-in-cell (PIC) simulations are essential for studying kinetic-scale plasma behaviour in astrophysical and laboratory environments, where processes like magnetic reconnection and collisionless shocks can accelerate particles to extreme energies. We present the recent advancements in the semi-implicit iPIC3D code, which now supports GPU computation using CUDA (NVIDIA GPUs) and HIP (AMD GPUs). Comparisons with the CPU version of the code show a factor of 30 improvement in computational speed. We have integrated the exact energy-conserving semi-implicit method and the relativistic semi-implicit method that ensures energy conservation up to machine precision. This is essential to avoid artificial energy growth in the system over long time scales and obtain physically viable results. Furthermore, the code has undergone several methodological improvements that further boost its speedup by 10x. Owing to the implicit nature of the code and the recent algorithmic advancements, we are on course for exascale simulations of relativistic and nonrelativistic plasma, enabling unprecedented spatial resolution and temporal duration.