Description

Graphics processing units (GPUs) have become a very powerful tool in high performance computing, offering significant parallelization in an oftentimes power efficient and cost effective manner. Certain types of computational problems are more compatible with the thread-based parallelization offered by GPU devices, and in this project we are working on accelerating simulations of particle-laden turbulence. In many applications, very large numbers of Lagrangian particles --- which can represent things like cloud droplets, sand grains, atmospheric aerosols, or contaminant mass --- are transported according to a turbulent flow being simulated at the same time. While traditional, MPI-based parallization of the flow solver are retained, calculation of the particle trajectories and fate is ported to multiple GPU devices, yielding significant performance gain and allowing for larger numbers of particles to be represented than could otherwise be possible. This project uses NVIDIA GPU devices and is based on the CUDA programming language.

For more information:

• Sweet, J., D.H. Richter, & D. Thain (2018). GPU acceleration of Eulerian-Lagrangian particle-laden turbulent flow simulations. International Journal for Multiphase Flow, 99, pp 437-445

1. Accelerate particle-laden flow simulations using GPUs
2. Quantify performance and identify the simulation regimes where this technique is beneficial
3. Explore further GPU-based enhancements, such as multiresolution time stepping and Lagrangian discretization of continuous fields