Toward Exa-Scale Computing in CAM-SE
Wednesday, 17 December 2014
According to the top500 website, supercomputers are projected to reach exa-flop scales within the next five years, enabling climate simulations with lateral resolutions that approach or exceed the hydrostatic barrier. This represents a regime-shift away from the familiar climate modeling techniques of the last 40 years into the nonhydrostatic regime where vertical acceleration, gravitational waves, and steep orography must be carefully modeled. The CAM-SE atmospheric dynamical-core is well positioned to take advantage of exa-flop supercomputers due to it near-optimal parallel scalability. However, scalability is merely a prerequisite and does not guarantee success. In order to perform accurate simulations, CAM-SE will require an array of new technologies including: nonhydrostatic dynamical solvers, greatly increased vertical accuracy, improved variable-resolution grid capabilities, and the integration of scale-aware physical parameterizations. In this talk, we will discuss the need for these new technologies and present our progress to date in their implementation.