High-Performance Weather Satellite Data Processing and Forecasting Model Advancement at SSEC using Accelerator Technology Current Status and Ongoing Endeavor

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Thursday, 8 January 2015: 8:30 AM
128AB (Phoenix Convention Center - West and North Buildings)
Allen Huang, CIMSS/Univ. of Wisconsin, Madison, WI; and B. Huang, J. Mielikainen, and M. Huang

Since 2009 a team at SSEC has devoted their ingenuity to leveraging high-performance parallel computing technology of NVIDIA GPUs to advance their satellite data applications. The specific areas of investigation are; 1) hyper-spectral data compression, 2) radiative transfer modeling for the Infrared Atmospheric Sounding Radiometer (IASI), 3) tsunami propagation modeling and, most recently, 4) weather prediction modeling. In 2014, the SSEC team, located at University of Wisconsin-Madison, has been selected as one of the new Intel Parallel Computing Center (IPCC) and has become a new NVIDIA CUDA Research Center (CRC) as well.

In this presentation we review the successful implementation of a GPU-based high-performance hyperspectral sounder radiative transfer model running on NVIDIA GPUs via CUDA (Compute Unified Device Architecture). We continue with a review of the progress made so far in the development of a GPU-based high-performance Weather Research Forecasting (WRF) model and demonstrate the design of a complete end-to-end GPU-CUDA WRF version, which could deliver a performance estimated to be in the range of 20X to 50X speedup with respect to a single, modern CPU core.

We conclude by reviewing our recent efforts in the use of Intel MIC Xeon Phi to advance WRF acceleration. Our initial results of a MIC Xeon implementation of WRF Thompson microphysics and TEMF planetary boundary layer scheme are outlined. We discuss our longer term plan to optimize the key schemes of microphysics, radiative transfer and planetary boundary physics components, and the dynamics code, starting from the computationally most time-consuming, namely the advection part.