APS Logo

HiPACE++: a portable, scalable 3D quasistatic particle-in-cell code

ORAL

Abstract

Modeling plasma wakefield accelerators is a computationally challenging task. Using cost-reducing algorithms like the quasistatic approximation allows for efficient modeling of demanding plasma wakefield accelerator scenarios. In this work, we present the performance-portable, 3D quasistatic particle-in-cell code HiPACE++. It adopts modern HPC practices like a performance-portability layer, continuous integration, standard I/O formats, and is open-source (https://github.com/Hi-PACE/hipace). Owing to careful memory management within the quasistatic algorithm, it demonstrates orders of magnitude speed-up on modern GPU-equipped supercomputers compared to its CPU-only predecessor HiPACE. We present a novel quasistatic pipeline algorithm, based on a temporal domain decomposition, which provides near-ideal scaling up to hundreds of GPUs. HiPACE++ enables efficient modeling of plasma wakefield accelerators both on state-of-the-art supercomputers as well as GPU-equipped laptops.

Presenters

  • Severin Diederichs

    Lawrence Berkeley National Laboratory, DESY, Lawrence Berkeley National Laboratory

Authors

  • Severin Diederichs

    Lawrence Berkeley National Laboratory, DESY, Lawrence Berkeley National Laboratory

  • Carlo Benedetti

    Lawrence Berkeley National Laboratory

  • Axel Huebl

    Lawrence Berkeley National Laboratory

  • Remi Lehe

    Lawrence Berkeley National Laboratory

  • Andrew Myers

    Lawrence Berkeley National Laboratory

  • Alexander Sinn

    DESY

  • Jean-Luc Vay

    Lawrence Berkeley National Laboratory

  • Weiqun Zhang

    Lawrence Berkeley National Laboratory

  • Maxence Thevenet

    DESY