P1.17
Parallel GPGPU acceleration algorithms for radiative transfer calculations in the atmosphere
William F. Godoy, NASA/LaRC, Hampton, VA; and X. Liu
General purpose computing on graphics processing units (GPGPU) is a recent programming technique that allows the highly parallel graphics processing unit (GPU) to perform calculations typically done by the central processing unit (CPU). In the present work, this technique is introduced to radiative transfer problems in the atmosphere. As an example, gas property calculations and 1D and 3D radiative transfer using the discrete ordinates method is solved and compared to radiances from satellite data. The methodology includes the treatment of flow control and parallelization of portions of a sequential algorithm, yielding to a large speed up in terms of computational times. For the practitioner, the underlying value of introducing a hybrid CPU/GPU algorithm is the potential application to current methods of solution to the different aspects of the overall problem (e.g. Monte Carlo, discrete ordinates, spherical harmonics, gas-particle property calculation, surface exchange) at the expense of a minimal learning curve.
Poster Session 1, Satellite Research and Algorithm Development in Meteorology - Posters
Monday, 27 September 2010, 5:30 PM-7:30 PM, ABC Pre-Function
Previous paper