9005: Progress of GPU-Parallel Numerical Weather Prediction and Climates Models
GTC Japan 2014
2014年7月16日
Stan Posey
Program Manager, HPC Solutions for Earth System Modeling
NVIDIA Corporation
Computational efficiency and increased performance per energy cost are key drivers behind scientific decisions to develop numerical weather prediction and climate models with GPU acceleration. Manageable compute costs with the growing modeling trends toward higher resolution, and with more and larger ensembles, may only be possible in the future with accelerated computing.
This presentation will examine the current state of GPU-parallel atmospheric and ocean models under development for operational prediction and observational research. GPU development strategies and application performance results will be given that compare the latest conventional CPUs both with and without GPU acceleration based on the Kepler architectural features from NVIDIA. Finally a forward looking discussion will be provided on relevant software tools and programmability requirements to achieve further HPC gains for the Earth system model community.