Aim: WP4 develops benchmarking strategies to gauge code efficiency and energy consumption for regional and global NWP applications on heterogeneous hardware architectures at scale. Moreover, to close the gap between the existing solutions in today’s NWP and climate models and canonical Weather & Climate Dwarfs, several steps are required.
Approach and methodology: The link is made by firstly setting up reference configurations of limited-area (LAM) and global model versions. Secondly, the LAM context requires a lateral boundary condition scheme (LBC) to be added, to enable the use of newly developed Weather & Climate Dwarfs that are building on the novel data structure framework developed in WP1. Finally, several alternative Weather & Climate Dwarfs will be tested in the global and the limited-area NWP context for specific hardware/accelerator options, while their performance is evaluated, in particular in view of their energy efficiency and running cost. For this purpose, WP4 will establish NWP model performance simulators for benchmarking including simulator architecture and design, simulator implementation and integration of the simulator with performance and energy models for selected hardware and applications. WP4 will also project the implications on the external workflow for selected LAM applications in the context of uncertainty estimation as planned in future NWP services.