The traditional Soble edge detection algorithms for optimization and implementation which were designed for common processor such as CPU, DSP and FPGA, could not be effectively applied on Graphics Processor Unit (GPU). A fast Sobel edge detection algorithm is presented based on NVIDA’s GPU which support Compute Unified Device Architecture (CUDA). On the basis of the parallel architecture and hardware characteristic of GPU, the fast algorithm introduces three methods to improve the implementation performance: Texture Storage technology optimizes the data storage structure, multiple point access technology improves the data access efficiency, and symmetry computation technology reduces the computation complex. The experiment result shows that GPU can effectively implement the fast algorithm and processing speed of 8-bit 4 096×4 096 pictures can be up to 190 fps, which is 122 times faster than CPU-based implementation.