A primary goal of the Next-Generation Global Weather Prediction System (NGGPS) program is to develop the Nation's next global weather prediction model to be used by the National Weather Service over the next five years. As part of this effort, five dynamical cores were evaluated in 2015, and two models were chosen for further testing: NOAA GFDL's Finite-Volume version 3 (FV-3) and the Model Prediction Across Scales (MPAS) from NCAR. While the Non-hydrostatic Icosahedral Model (NIM) was not selected for further evaluation, it was the only model able to run on traditional CPU, and fine-grain GPU and MIC architectures, and demonstrates the potential to run high-resolution global prediction models on fine-grain systems in the future. Using the NIM as a guide, efforts are underway to parallelize the MPAS and FV-3 models for fine-grain architectures. We will briefly describe the status of these efforts. We will also present performance comparisons of the NIM based on four generations of CPU and GPU chips (2010-2014), and one generation of Intel MIC, the Intel Knights Corner chip (2013). Using the 2014 performance results and published list prices, we will show cost-benefit results for single device, node, and full system configurations.
This publication was presented at the following:
Authors who have authored or contributed to this publication.