Abstract: Digital speckle correlation method has the advantages of low demand of the measurement environment, overall non-contact measure, however, the algorithm efficiency has been one of the bottlenecks limiting its development. The graphics processing unit (GPU) has the natural parallelism, GPU highperformance computing brings great efficiency on computer image processing. By programming with the compute unified device architecture (CUDA) platform, the GPU high-performance parallel processing was applied for the traditional digital speckle point-by-point search algorithm, cross search algorithm and genetic algorithm.Comparing with the traditional method, the experimental results show that the efficiency of these 3 methods improve by 20,8,31 times respectively for 150150 pixels speckle image, while by 183,33,44 times respectively for 500500 pixels speckle images and by 424,116,44 times respectively for 1 0001 000 pixels speckle images.