The adaptive optical system based on stochastic parallel gradient descent algorithm (SPGD) compensates wavefront aberration by direct optimization of a system performance metric. The convergence rate is the key factor that limits the SPGD algorithm application in practical adaptive optics system. On the basis of analysis of SPGD algorithm in which the concurrency is embedded, a pipeline single instruction and multipe data (SIMD) architecture parallel real-time wavefront processor is deslgned and implementd in which field-programmable gate array (FPGA) and digital signal processing (DSP) are introduced. The optimal mapping of the control algorithm from representative layer to the architecture is realized. The wavefront processor is applied to a laser beam cleanup adaptive optical system to control the deformable mirror and fast steering mirror. The experimental results show that real-time parallel wavefront processor based on SPGD algorithm can compensate the wavefront aberration and tilt excursion of the laser beam involved effectively and that the convergence rate is enhanced enormously.