Semi-Lagrangian 4d, 5d, and 6d kinetic plasma simulation on large scale GPU equipped supercomputer

TL;DR

This paper introduces efficient semi-Lagrangian discontinuous Galerkin methods for high-dimensional kinetic plasma simulations on large-scale GPU supercomputers, achieving high performance and scalability.

Contribution

It presents a novel implementation of 4d, 5d, and 6d plasma simulations on GPUs, demonstrating significant performance and scalability improvements.

Findings

Single node performance exceeds 2 TB/s memory bandwidth.

Achieves parallel efficiency of 30% to 67% on 1536 GPUs.

Successfully scales high-dimensional plasma simulations on large GPU clusters.

Abstract

Running kinetic plasma physics simulations using grid-based solvers is very demanding both in terms of memory as well as computational cost. This is primarily due to the up to six-dimensional phase space and the associated unfavorable scaling of the computational cost as a function of grid spacing (often termed the curse of dimensionality). In this paper, we present 4d, 5d, and 6d simulations of the Vlasov--Poisson equation with a split-step semi-Lagrangian discontinuous Galerkin scheme on graphic processing units (GPUs). The local communication pattern of this method allows an efficient implementation on large-scale GPU-based systems and emphasizes the importance of considering algorithmic and high-performance computing aspects in unison. We demonstrate a single node performance above 2 TB/s effective memory bandwidth (on a node with 4 A100 GPUs) and show excellent scaling (parallel efficiency between 30% and 67%) for up to 1536 A100 GPUs on JUWELS Booster.