You are here

Development of a GPU-based High Performance Community Radiative Transfer Model

Award Information
Agency: Department of Commerce
Branch: National Oceanic and Atmospheric Administration
Contract: WC133R-11-CN-152
Agency Tracking Number: 11-10
Amount: $94,946.00
Phase: Phase I
Program: SBIR
Solicitation Topic Code: 8.3.5D
Solicitation Number: 1
Solicitation Year: 2011
Award Year: 2011
Award Start Date (Proposal Award Date): 2011-09-07
Award End Date (Contract End Date): 2012-03-06
Small Business Information
14 La Pointe Terrace
Madison, WI 53719-
United States
DUNS: 135228216
HUBZone Owned: No
Woman Owned: No
Socially and Economically Disadvantaged: No
Principal Investigator
 Allen Huang
 Principal Scientist
 (608) 628-2468
Business Contact
 Allen Huang
Title: Managing Director
Phone: (608) 628-2468
Research Institution

Computation of the radiative transfer model for a hyperspectral sounder with thousands of spectral channels is very time-consuming. Consequently, operational data assimilation systems can assimilate only a few hundred channels. The radiative transfer model is very suitable for GPU implementation to take advantage of GPU massively parallel computing capability, where radiances at various channels can be calculated simultaneously. Our recent paper demonstrated that a GPU-based radiative transfer model for the IASI sounder with 8461 channels could be 1523x faster than its original single-threaded CPU version. It means that one day's amount of 1,296,000 IASI spectra can be calculated within 15 minutes on a low-cost GPU computer, whereas the original CPU code would take more than 15 days in the host PC. The innovation of this work is our development of heterogeneous pipelining and asynchronous transfer between CPUs and GPUs for the significant speedup.

Inspired by this success, we propose to develop a GPU-based high-performance Community Radiative Transfer Modeling (CRTM) for NOAA. During Phase 1, we will demonstrate the feasibility of further enhancing our GPU techniques to allow simultaneous computation of multiple hyperspectral radiance spectra on GPUs. We expect to double the speedup (~3000x) for use in data assimilation and atmospheric soundings.

* Information listed above is at the time of submission. *

US Flag An Official Website of the United States Government