2014 AMS Annual Meeting

Continued advancements in numerical weather prediction are tied to running models at higher temporal and spatial scales.  Next-generation global models are being designed to run at 3KM resolution, with North-American models proposed to run at 1KM or finer resolution.  An estimated 10 to 100 fold increase in high performance computing is needed to support development, testing, and running these models, and related ensemble and data assimilation systems, at these resolutions.  HPC resources will be needed to develop, test, and tune the models as well as run them in highly reliable operational environments so products and grids can be delivered in a timely manner to downstream users in the government, commercial and public sectors.  

The emphasis of this town-hall is on the HPC necessary for running weather models at sustained teraflop or even petaflop performance in production and development environments. We will discuss issues including code scalability, choice of accelerator technologies (MIC vs. GPGPU), managing I/O, and data locality for analysis and post-processing.

For additional information, please contact Brian Etherton (e-mail: [email protected]).