Evaluating Computational Performance for the Weather Service's Next Generation Global Prediction System

As part of the National Weather Service's five year Research to Operations Initiative, and in cooperation with NOAA's Office of Atmospheric Research, NOAA is designing, developing, and implementing the Next Generation Global Prediction System that will run operationally as a fully coupled atmosphere-ocean-wave-land-aerosol system in 2020. A major thrust of the NGGPS project is upgrading the current Global Forecast System running at NCEP with a new dynamical core designed to take full advantage of successive generations of High Performance Computing systems that will verge towards exascale (10¹⁸ floating point operations per second) capability over the 10 to 15 year life span of the NGGPS. Work has begun to identify and characterize the scaling challenges for operational NWP at these scales, and a comprehensive testing regimen is underway to evaluate both forecast quality and computational performance of a set of candidate model dynamical cores from leading developer groups in NOAA, the Navy, and NCAR. To provide additional perspective, we also test an operational dynamical core from the European Centre for Medium Range Weather Forecasts. Benchmarks are conducted on currently vanguard HPC systems comprising more than 100K processors. This is in anticipation of even greater capability from future HPC systems. In this presentation, we will describe the NGGPS planning and testing with results to date from the candidate model evaluations.