To take full advantage of these new satellite assets, and to be able to provide the real-time information for time-critical, decision-making needed to save lives, protect property, and mitigate risks and loss, we must get ready for their utilization. This means full leverage of high-performance and energy-efficient computing technology now becoming available to us. High-speed, low-cost and low-energy-consumption Graphics Processing Units (GPUs) are emerging as the backbone of the next generation of low-cost, small footprint, super-computing engines. Currently the top-performing super computers use GPU technology to achieve both cost and energy advantages over CPU configurations of equivalent processing power. The combined features of general-purpose GPU supercomputing — i.e. high parallelism, high memory bandwidth (102 GB per GPU), low cost (a few thousands of dollars), and compact size (1U) — are what make a GPU-based personal desktop supercomputer an appealing alternative to a massively parallel system made up of commodity CPUs (e.g. Beowulf clusters).
This paper starts with a review of our successful implementation of a GPU-based high-performance IASI radiative transfer model running on NVIDIA GPUs via CUDA (Compute Unified Device Architecture), NVIDIA's parallel computing architecture for massively multi-threaded parallel computation. The paper continues with a review of the progress made so far in developing a GPU-based high-performance implementation of the Weather Research and Forecasting (WRF) model and a review of the status of a similar implementation of the ultra spectral sounder component of Community Radiative Transfer Model (CRTM). We conclude 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, these 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 throughput of future geosynchronous earth remote sensing systems such as GOES-R, GMW, MTG, FY-4, et al. for real-time product generation, numerical weather prediction and data assimilation.