Flexible CPU-GPU computational plasma applications with particles and fluids

Theoretical computing speeds have increased dramatically in the last decade with chip advances, including Graphics Processor Units (GPUs).In practice, however, these gains have not been realized, largely due to the accompanying high development costs due to the need for multiple code bases to take advantage of the various chips and their variants.The proposed project addresses these issues from both sustainability and in providing the particular computational capability needed by the defense community in the area of plasma modeling by both fluid and particle-in-cell methods.Phase I prototyped communication with devices, developed algorithms for fluid dynamics and particle-in-cell methods, and developed flexible patterns that support multiple chip architectures at minimal cost.Phase II will build on these results to deliver a flexible computational application able to compute across both the CPU and any number of devices, with runtime detection and automatic determination of the best algorithm, to switch automatically between fluid and particle algorithms as the collisionality changes, and to be developed by a migration path so that capability is not lost.Government benefits include advancing research in high-energy density plasmas.Commercial benefits include improved modeling of plasma discharges, satellite charging, and plasma medicine for design.