This article is cited in 2 papers

Informatics

Development and realization of algorithm of coagulation calculation in mixed-phase cloud model using CUDA technology

N. O. Raba

St. Petersburg State University, Department of Mathematics and Mechanics

Abstract: Numerical modeling is one of the most effective way to investigate convective clouds. Mixed-phase cloud models with spectral microphysics are usually applied for more accurate forecast. Computer resources are mainly spent on microphysical processes calculations (especially on coagulation process). One of the methods of computations speed-up is parallel computing. Modern central processors have only 2–6 cores in contrast to a few hundreds of cores in video cards. Special parallel algorithms are required for such numbers of cores. In the paper a coagulation calculation algorithm for a mixed-phase cloud model is described. This algorithm is modified version of the algorithm for liquid phase cloud model. A method of parallelization of this algorithm for computation on video card using CUDA technology is also proposed. A way of avoid simultaneous access to memory cells from different threads is demonstrated. The testing shows significant speeding-up of coagulation calculation.

Keywords: parallel computing, CUDA, GPGPU, coagulation, coalescence, spectral microphysics.

UDC: 004.272.23+519.681.3


Accepted: May 19, 2011