1. Institute of Geology and Geophysics, China Academy of Sciences, Beijing City, 100029, China;
2. College of Earth Sciences, Northeast Petroleum University, Daqing City, Heilongjiang Province, 163318, China
Abstract：SRME method to predict surface multiples require a lot of space convolution operation,exactly speaking,the vast calculation were from the matrix multiplication operation in each frequency component.A better calculation scheme can be expressed as below:assigning the vast matrix multiplication operation to GPU for each frequency component,so the efficiency for surface multiple prediction was greatly raised,the other calculation were conducted by CPU,as a result the suppression of surface multiple was conducted by the collaborative parallel calculation from GPU and CPU,in this way the calculation efficiency for SRME was greatly improved.In this paper prediction of surface multiple of the SMAART model by GPU and CPU were respectively conducted,the result shows that application of GPU calculation technique could greatly accelerate the prediction of surface multiple.The minimum energy self-adapting subtracting method was used to suppress the surface multiples which were predicted by GPU and CPU,the result indicates that the calculation result from GPU technique could totally keep the calculation precision of CPU technique.
