Relativistic Hydrodynamics on Graphic Cards

TL;DR

This paper demonstrates how to significantly accelerate relativistic hydrodynamics simulations using GPUs, enabling more efficient event-by-event analyses in high-energy nuclear physics.

Contribution

The authors redesigned the SHASTA algorithm with OpenCL to run efficiently on GPUs, achieving a 160-fold speedup over traditional CPU implementations.

Findings

Hydrodynamic calculations are accelerated by a factor of 160.

GPU implementation enables event-by-event simulations.

Improved computational efficiency benefits high-energy nuclear physics research.

Abstract

We show how to accelerate relativistic hydrodynamics simulations using graphic cards (graphic processing units, GPUs). These improvements are of highest relevance e.g. to the field of high-energetic nucleus-nucleus collisions at RHIC and LHC where (ideal and dissipative) relativistic hydrodynamics is used to calculate the evolution of hot and dense QCD matter. The results reported here are based on the Sharp And Smooth Transport Algorithm (SHASTA), which is employed in many hydrodynamical models and hybrid simulation packages, e.g. the Ultrarelativistic Quantum Molecular Dynamics model (UrQMD). We have redesigned the SHASTA using the OpenCL computing framework to work on accelerators like graphic processing units (GPUs) as well as on multi-core processors. With the redesign of the algorithm the hydrodynamic calculations have been accelerated by a factor 160 allowing for event-by-event calculations and better statistics in hybrid calculations.