Numerical Simulation Of Groundwater Flow Based On CUDA

The use of numerical simulation in analyzing trends of groundwater, can effectivelydeal the problem of the evaluation and management of water resources. At present, therapid progress of groundwater numerical simulation software, more and more parallelprogram makes the operation speed of groundwater numerical model has improvedgreatly. It’s so needed that programming a parallel numerical groundwater modelingsoftware which is low cost and high efficiency for the researchers.Based on the unified computing architecture platform which NVIDIA launched（AGENERAL-PURPOSE PARALLEL COMPUTING PLATFORM ANDPROGRAMMING MODEL, Compute Unified Device Architecture, referred to asCUDA）,combine with the new package GPCG developed for MODFLOW,we realizedthe parallel of MODFLOW on GPU. Research results show that the program canquickly achieve parallelism has broad application prospects with the CUDAprogramming.Through the research of CUDA platform, we recognized the advantages in parallelcomputing on CUDA. We established three different scales ideal model (1,000,000grids,2,000,000grids,4,000,000grids) for testing the calculation accuracy and efficiency ofMODFLOW improved and got the results of overall speedup about2.5times,iterative solver part about11times speedup. It proved that MODFLOW based-on theCUDA platform is feasible.We applied the program in the North China Plain, Daxing karst water on differentscale numerical simulation to verify the credibility of the reform program the results,we got the result that the speedup is about2.16times. Data analysis showed that theimproved MODFLOW program can effectively reduce the share of part-time modeliterations overall running time of proportion; comparing the results of different scalemodels showed that with increase in the number grid, CUDA platform can solve themore obvious advantages, the speedup higher.This article is based on research GPCG package of MODFLOW on CUDA,itmakes the program can be used on laptops, computers and other devices with GPU and provides a suggestion for achieving low-cost, high-efficient parallel computing.