source: Exascale Computing Project
In this interview, award winning scientist Mark Taylor at Sandia National Laboratories’ Center for Computing Research talks about the use of exascale-class supercomputers – to be delivered to three U.S. Department of Energy national labs in 2021 – for large-scale weather and water resource forecasting.
Taylor is chief computational scientist for the DOE’s Energy Exascale Earth System Model (E3SM), a sub-project within DOE’s Exascale Computing Project (ECP). As PI, Taylor along with a team of researchers are working to enhance simulations extending ahead across multiple decades of the water cycle and other processes around precipitation.
The project team is bringing together E3SM’s many components, including models for atmosphere, land, ocean, ice, sea ice and land ice, within a single, massive software framework. The core need is improved resolution: current standards for climate change assessment are at about 100-kilometer resolution, but the goal of E3SM is to reduce that to 1 kilometer, requiring 1008 more compute power. In preparation for exascale’s arrival, Taylor said the team is using DOE’s current GPU-based pre-exascale systems to evaluate which applications are best suited for porting to exascale.
For example, E3SM researchers ported the Multiphysics Modeling Framework’s “nonhydrostatic atmosphere dynamical core” to run on Summit. Originally written in Fortran, the core was rewritten in C++/Kokkos for on-node parallelism, enabling optimization of Summit’s GPU (Nvidia V100) processing power.
This, in combination with advanced algorithms, resulted in record setting performance for a global cloud resolving dynamical core, according to E3SM, and “will enable some of the first decade-long cloud-resolving climate simulations. Such simulations are essential for scientists to fully address the problems arising from approximations in cloud systems used in traditional climate models. In the exascale era, these types of simulations will remove one of the major sources of climate prediction uncertainty.”
The ultimate objective: improved long-term forecasting of rain, of flooding and of droughts, along with anticipated rises in the sea level. If the Exascale Earth System Model delivers on its goals, people, communities and countries will be better able to adjust to anticipated safety threats and food chain disruptions brought on by climate change.
