Optimizing Vector Particle-In-Cell (VPIC) for Memory Constrained Systems Using Half-Precision

Student: Nigel P. Tan (University of Tennessee)
Supervisor: Michela Taufer (University of Tennessee, Knoxville)

Abstract: Vector Particle-In-Cell (VPIC) is one of the fastest plasma simulation codes in the world, with particle numbers ranging from one trillion on the first petascale system, Roadrunner, to more recent 10 trillion particles on the Trinity supercomputer. Current memory systems limit VPIC simulations greatly as the maximum number of particles that can be simulated directly depends on the available memory. In this work we present a suite of VPIC optimizations (i.e., particle weight storage and half-precision position storage optimizations) that enable significant increases to the number of particles. We assess the optimizations' impact on a GPU-accelerated Power9 system. We show how our optimizations enable a 40% increase in the number of particles simulated in VPIC.

ACM-SRC Semi-Finalist: yes

Poster: PDF
Poster Summary: PDF


Back to Poster Archive Listing