TJ46.1 The Development of GPU-based High-Performance Models in Support of Real Time Weather and Environmental Applications

Thursday, 10 January 2013: 11:00 AM
Room 9C (Austin Convention Center)
Allen Huang, CIMSS/Univ. of Wisconsin, Madison, WI; and B. Huang and J. Mielikainen

In recent years the General-Purposed Graphics Processing Unit (GPGPU) with thousands of compute cores, and high memory bandwidth has made a low-cost personal GPU supercomputer an appealing alternative to a massively parallel system made up of commodity CPUs. For example, the recent commercially available NVIDIA Kepler GTX690 achieves more than 5,000 GFlops but cost less than $1,000. In this paper we'll show several examples of GPU-based models for the accelerating of satellite data in time critical processing and applications.

This paper starts with a review of our recent successful implementation of a GPU-based high-performance hyperspectral sounder radiative transfer model running on NVIDIA GPUs via CUDA (Compute Unified Device Architecture), the compute engine in NVIDIA GPUs designed for massively multi-threaded parallel computation. The paper follows by the review of the progress been made so far for the development of a GPU-based high-performance (WRF) model and concludes with an emphasis on the design and demonstration of a GPU-based high-performance computing infrastructure to be developed for the next generation of satellite sensor systems, such as hyper-spectral imagers and ultra-spectral sounders, to be flown aboard both polar-orbiting and geostationary platforms. This is our first step toward the development of a GPU-based high-performance computing infrastructure to support sustained processing and applications of current and future LEO and GEO earth remote sensing systems for real-time product generation and numerical weather prediction and data assimilation and time-critical applications.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner