Adaptive optics systems with larger size require more subapertures in the Shack-Hartmann wavefront sensor, which makes it harder to compute the wavefront gradient in real time. In order to measure local gradients in large-size adaptive optics systems, an architecture based on cross correlation is proposed. Using the Gaussian spot as the reference template, the correlation between subaperture spots is calculated by means of two cascading filters, and then the local wavefront gradient of each subaperture is determined. Compared with the existing parallel processing approach in correlating-Hartmann sensor, the cost of our architecture will not increase greatly as subapertures multiplies, which makes it more suitable for adaptive optics systems of extreme size to retrieve gradients. Experimental results indicate that when implemented in Field Programmable Gate Array (FPGA) at 27 MHz, our design can obtain gradients of all subapertures in 8×8 Shack-Hartmann with latency no more than 1.2 μs, while the RMS error is less than 0.02 pixels, and the maximum error does not exceed 0.04 pixels. It can satisfy the requirements of low cost, real time and high accuracy for gradient detection in adaptive optics systems with large-size aperture.