Enhancing Efficiency of the RRTMG Radiation Code with GPU and MIC Approaches for Numerical Weather Prediction Models
Atmospheric and Environmental Research, Lexington, Massachusetts
14th Conference on Cloud Physics/14th Conference on Atmospheric Radiation/Anthony Slingo Symposium, 2014
@article{iacono2014enhancing,
title={Enhancing Efficiency of the RRTMG Radiation Code with GPU and MIC Approaches for Numerical Weather Prediction Models},
author={Iacono, Michael J. and Berthiaume, David},
year={2014}
}
Radiative transfer (RT) calculations are among the most computationally expensive components of global and regional weather and climate models, and radiation codes are therefore ideal candidates for applying techniques to improve the overall efficiency of such models. In many general circulation models (GCMs), a physically based radiation calculation can require as much as 30-50 percent of the total GCM execution time. This work examines two methods of enhancing the efficiency of the widely used RRTMG radiation codes by employing computing hardware designed for code acceleration such as graphics processing units (GPUs) and the Intel Xeon Many Integrated Core (MIC) architecture rather than traditional multi-core central processing units (CPUs). In some dynamical models, the computational expense of RT in GCMs has been improved by reducing the spatial or temporal resolution of the radiation calculations, but this approach may degrade the calculated radiative fluxes and heating rates and possibly impair the simulation. One motivation for using high performance computing techniques is to allow RT calculations to be performed at appropriate spatial and temporal resolutions as well as making additional computational time available to improve other dynamical model components.
September 5, 2014 by hgpu