THE RAPID RADIATIVE TRANSFER MODEL RRTMG, DEVELOPED AT ATMOSPHERIC AND ENVIRONMENTAL RESEARCH, INC. (AER), WILL SOON BE THE OPERATIONAL RADIATION CODE IN THE GEOS-5 ATMOSPHERIC MODEL. THE PROPOSERS, THE DEVELOPERS OF RRTMG, HAVE BEEN COLLABORATING WITH NASA SCIENTISTS TASKED BY GMAO WITH IMPLEMENTING THIS GEOS-5 UPGRADE, TO ENSURE THAT THE FULL BENEFIT OF RRTMG'S CAPABILITIES IS ATTAINED BY GEOS-5. THE APPROACH USED BY RRTMG TO COMPUTE BROADBAND THERMAL AND SOLAR RADIATIVE FLUXES AND HEATING RATES IS FOUNDED ON THE CORRELATED-K-METHOD, ENHANCED BY NUMEROUS INNOVATIVE ALGORITHMIC ADVANCES, ALLOWING THE CODE TO ACHIEVE AN IMPRESSIVE COMBINATION OF ACCURACY AND COMPUTATIONAL SPEED. IN PARTICULAR, RRTMG IS AN ACCELERATED VERSION OF THE AER CODE RRTMG; A GREAT DEAL OF DEVELOPMENT HAS ALLOWED RRTMG TO ATTAIN SUBSTANTIAL IMPROVEMENTS IN COMPUTATIONAL EFFICIENCY OVER RRTM WITH ONLY A SLIGHT DECREASE IN ACCURACY. (THE `G IN RRTMG REFERS TO THE GCM-READY STATE OF THE CODE.) DESPITE THESE SPEED IMPROVEMENTS, THE FRACTION OF TIME SPENT COMPUTING RADIATIVE FLUXES IN A GCM LIKE GEOS-5 REMAINS ~30% OF THE TOTAL COMPUTATIONAL TIME. THE OVERALL COMPUTATIONAL EXPENSE CONSTITUTES A LIMITATION ON A GCM'S PREDICTIVE ABILITY IF IT BECOMES AN IMPEDIMENT TO ADDING NEW PHYSICS TO OR INCREASING THE HORIZONTAL AND VERTICAL RESOLUTION OF THE MODEL. THE COMPUTATIONAL EFFICIENCY OF RRTMG IS UNLIKELY TO BE SUBSTANTIALLY IMPROVED BY FURTHER CPU-BASED SOFTWARE ENHANCEMENTS. EVOLUTION OF THE CODE TO GPU-BASED TECHNOLOGY WILL PERMIT A SIGNIFICANT SPEED UPGRADE, THEREBY PERMITTING ALGORITHMIC ENHANCEMENTS THAT IMPROVE THE PREDICTIVE ABILITY OF THE GCM. WE PROPOSE TO CREATE A VERSION OF THE RRTMG RADIATION CODE DESIGNED TO RUN EFFICIENTLY IN A GPU ENVIRONMENT. THIS EFFORT WILL FOCUS ON THE RRTMG IMPLEMENTATION IN GEOS-5; WE WILL WORK CLOSELY WITH DR. MAX SUAREZ, THE SCIENTIST AT GMAO RESPONSIBLE FOR THE RADIATION PACKAGE IN GEOS-5. RRTMG HAS AN INTERNAL PSEUDO-SPECTRAL VECTOR OF LENGTH ~100 THAT, WHEN COMBINED MUCH GREATER LENGTH OF THE GLOBAL HORIZONTAL GRID VECTOR FROM WHICH THE RADIATION CODE IS CALLED IN GEOS-5, MAKES RRTMG/GEOS-5 PARTICULARLY SUITED TO ACHIEVING A SIGNIFICANT SPEED IMPROVEMENT THROUGH GPU TECHNOLOGY. THIS LARGE NUMBER OF INDEPENDENT CASES WILL ALLOW US TO TAKE FULL ADVANTAGE OF THE COMPUTATIONAL POWER OF THE LATEST GPU S, ENSURING THAT ALL THREAD CORES IN THE GPU REMAIN ACTIVE, A KEY CRITERION FOR OBTAINING SIGNIFICANT SPEEDUP. THE CUDA FORTRAN COMPILER DEVELOPED BY PGI AND NVIDIA WILL ALLOW US TO CONSTRUCT THIS PARALLEL IMPLEMENTATION ON THE GPU WHILE REMAINING IN THE FORTRAN LANGUAGE. THIS IMPLEMENTATION WILL SCALE VERY WELL ACROSS VARIOUS CUDA SUPPORTED GPU S SUCH AS THE RECENTLY RELEASED FERMI NVIDIA CARDS INCLUDING THE TESLA COMPUTING SYSTEMS. THE CODE THAT WILL BE DEVELOPED WILL BE SUBJECTED TO EXTENSIVE VALIDATION, EXPLOITING OUR UNPARALLELED KNOWLEDGE OF RRTMG'S STRUCTURE AND FUNCTIONALITY AS WELL AS OUR VAST EXPERIENCE EVALUATING RRTMG'S RADIATION CALCULATIONS RELATIVE TO THOSE FROM AER'S MORE DETAILED RADIATION CODES (RRTM AND THE LINE-BY-LINE RADIATIVE TRANSFER MODEL). WE WILL ALSO COLLABORATE WITH DR. SUAREZ TO EVALUATE THE GPU-VERSION OF RRTMG/GEOS-5 IN APPLICATIONS OF THE FULL DYNAMICAL MODEL. THE END RESULT OF THE PROPOSED WORK WILL BE A GPU-COMPATIBLE VERSION OF RRTMG FOR APPLICATION TO GEOS-5, THEREBY ALLOWING THE GCM TO EXECUTE MORE EFFICIENTLY AND/OR TO BE IMPROVED BY ADDING OTHER PHYSICS SOFTWARE PACKAGES THAT HAD HERETOFORE BEEN CONSIDERED TOO COMPUTATIONALLY EXPENSIVE. RRTMG IS ALSO THE OPERATIONAL RADIATION CODE IN OTHER DYNAMICAL MODELS, SUCH AS NCEP'S GLOBAL FORECAST SYSTEM AND CLIMATE FORECAST SYSTEM, NCAR'S COMMUNITY EARTH SYSTEM MODEL AND COMMUNITY ATMOSPHERE MODEL, AND ECMWF'S INTEGRATED FORECAST SYSTEM. WE HAVE WORKED CLOSELY WITH THESE ORGANIZATIONS TO IMPLEMENT RRTMG APPROPRIATELY IN EACH MODEL. THIS PROPOSED EFFORT IS LIKELY TO LEAD TO SIGNIFICANT ASSOCIATED BENEFITS FOR THESE OTHER GCM APPLICATIONS.