In this paper, a real-time on-line performance evaluation processor based on graphic processing unit (GPU) for adaptive
optics (AO) system is presented, aiming to monitor the 127-element AO system during its close-loop work by
quantifying its correction results, which can provide reference to improve the performance of the system. Since there is a
contradiction between the heavy computation burden and the real-time processing requirement, we modified operations
and algorithms to fit the CPU-GPU heterogeneous environment, in which GPU is used to handle the complex
computation but simple logicality, and CPU is assigned to undertake data transportation between internal storage and
video memory，as well as some small-scale computations. In the real-time processor, performance parameters to be
computed include peak-valley (PV) and root-mean-square (RMS) of near-field wavefront phase, point spread function
(PSF), full width half maximum (FWHM) of far-field image，modulation transfer function (MTF) and Strehl ratio (SR).
And the inputs are residual slopes obtained from Hartmann wavefront sensor of 127-element AO system. By
computation 4096 frames of parameters, the average rate by single precision is 4.11ms/frame.