A022-15
Recent performance improvements in DOE E3SM fully coupled model
Monday, 7 December 2020: 16:42
Virtual
Jean-Christophe Golaz1, Luke Van Roekel2, David Bader1, Andrew M. Bradley3, Yan Feng4, Oksana Guba5,6, Walter M Hannah1, Jayesh Krishna7, L. Ruby Leung8, Wuyin Lin9, Po-Lun Ma8, Mathew E Maltrud2, Andrew Roberts10, Qi Tang1, Mark A Taylor6, Milena Veneziani2, Jonathan D Wolfe2, Danquing Wu7, Shaocheng Xie1, Kai Zhang11, Xue Zheng1 and Tian Zhou12, (1)Lawrence Livermore National Laboratory, Livermore, CA, United States, (2)Los Alamos National Laboratory, Los Alamos, NM, United States, (3)Sandia National Laboratory, Albuquerque, NM, United States, (4)Argonne National Laboratory, Environmental Science Division, Argonne, IL, United States, (5)Lethbridge, AB, Canada, (6)Sandia National Laboratories, Albuquerque, NM, United States, (7)Argonne National Laboratory, Argonne, United States, (8)Pacific Northwest National Laboratory, Richland, WA, United States, (9)Brookhaven National Laboratory, Upton, NY, United States, (10)Naval Postgraduate School, Monterey, CA, United States, (11)Pacific Northwest National Lab, Richland, WA, United States, (12)Pacific Northwest National Laboratory, Atmospheric Sciences and Global Change Division, Richland, WA, United States
Abstract:
The first version of the Department of Energy’s Exascale Energy Earth System Model (E3SMv1) was released two years ago. Since then, considerable efforts have been dedicated to improving the model’s computational performance. These efforts yielded a nearly 2x performance gain for the standard resolution configuration (100 km atmosphere, 30-60 km ocean) of E3SMv2 compared to ESMv1. In the atmosphere, performance gains were achieved as a result of a new dynamical core with a semi-Lagrangian tracer transport scheme and a slightly coarser finite volume grid for physics calculations. A newly developed parallel IO library (SCORPIO) also yielded performance gains.
The quality of the simulated climate in E3SMv2 is generally comparable to E3SMv1 with some notable improvements. In particular, updates to ocean mesoscale eddy parameterizations led to reductions in regional sea-surface temperature (SST) biases as well as a more realistic Atlantic meridional overturning circulation (AMOC). In the atmosphere, cloud and precipitation biases were reduced through careful calibration of the existing cloud parameterizations. While simulations with E3SMv2 are run on traditional (CPU) HPC resources, the next version will target GPU-enabled architectures. We will review our path towards Exascale.
Work at LLNL was performed under the auspices of the U.S. DOE by Lawrence Livermore National Laboratory under contract No. DE-AC52-07NA27344.
