Adaptive affinity propagation clustering(adaptive Affinity propagation clustering, adAP), as a new clustering algorithm, does not need to specify the initial "exemplars" and the class number, which is effective to solve the problem of class number uncertainty in clustering. Then, as a result of the adAP is extremely time consuming, the larger the number of samples is, the slower the speed is. In order to improve the speed of the adAP, this paper realizes a parallel method, which is based on NVIDIA's Compute Unified Device Architecture (CUDA) and Matlab parallel computing toolbox. The experiment results show that the GPU-based parallel adAP method has a certain speedup effect, and it is more than 2 times faster than the serial execution. With the increase of the number of samples, the acceleration performance is getting better and better.