In 2015, Giulia Matrone proposed a nonlinear beamforming algorithm, Delay Multiply and Sum Beamforming (DMAS), which is superior to the Capon algorithm and Delay and Sum Beamforming (DAS) algorithm in resolution and contrast. The complexity of the DMAS algorithm is O(n²). In this paper, a parallel beamforming algorithm named PDMAS is proposed. The complexity of the PDMAS algorithm is O(n). In PMDAS algorithm, the channel operation is independent with each other and with good parallelism. At the same time, the software architecture of PDMAS algorithm is designed on GPU by using OpenCL programming language to test the performance of the proposed algorithm. The experiments are performed on the Precision T7810 workstation which contains two CPUs (Intel (R) Xeon (R) CPU E5-2620 v4) and AMD FirePro W7100 GPU respectively. The experimental result shows that both the PDMAS and the DMAS algorithm can't meet the real-time imaging requirements on the CPU. On the GPU, the frame rate of the DMAS algorithm is greatly improved, but still can't meet the real-time imaging requirements. On the other hand, the frame rate of PDMAS algorithm on the GPU can reach to 83fps, which means can fully meet the needs of real-time imaging. It can be concluded that the PDMAS algorithm has a better performance on GPU and fully meet the requirements of ultrasound real-time imaging systems.