OP-E3.2 The SP-E3SM

Owned by Ben Hillman
Last updated: Nov 08, 2018 by Renata McCoy

E3SM Session

Session ID: E3 (would also fit E5, E8)

                    

Poster TitleThe Super-Parameterized Energy Exascale Earth System Model (SP-E3SM)
AuthorsBen Hillman, Walter Hannah, Christopher Jones (Unlicensed), Matthew Norman, Sarat Sreepathi, Mark Taylor, David C. Bader, Ruby Leung
First AuthorBen Hillman
Session TypeE3SM Session
Session IDE3
Submission TypePresentation
GroupAtmosphere
Experiment
Poster Link




Abstract

The Energy Exascale Earth System Model (E3SM) Project is a state-of-the-science Earth system model that optimizes the use of DOE’s advanced supercomputers to meet the science needs of the future. A “super-parameterized” version of the E3SM (SP-E3SM) is under development, in which the moist physics parameterization suite is replaced with a cloud resolving model (CRM) in each grid column. In order to maximize computational performance, the CRM has been refactored to expose threading and offload work to GPU accelerators using a novel hybrid directives-based approach that maps a single set of directives to either openMP or openACC at compile time from a single source. This enables the use of potentially four compilers for GPU offloading: Cray, PGI, GNU, and XL. A new radiation code (RRTMGP) has also been implemented that is designed with improved parallelism in mind. Here, we present early simulations with the SP-E3SM and compare its performance against typical configurations of the E3SM. Our analysis emphasizes key features of the water cycle, including the diurnal cycle of precipitation and the distribution of rain rates, and reveals similarities with previous super-parameterized model development (mesoscale convective system activity, diurnal cycle fog, and a lack of shallow convection in spite of higher vertical resolution), but also some surprising sensitivities that have not been explored, including imprinting of the spectral element grid used in the dynamical core and differences in the behavior of convection that manifest as a widening of the precipitation distribution.