Compute unified device architecture (CUDA) is a software development platform that enables us to write and run general-purpose applications on the graphics processing unit (GPU). This paper presents a fast method for cone beam reconstruction using the CUDA-enabled GPU. The proposed method is accelerated by two techniques: (1) off-chip memory access reduction; and (2) memory latency hiding. We describe how these techniques can be incorporated into CUDA code. Experimental results show that the proposed method runs at 82% of the peak memory bandwidth, taking 5.6 seconds to reconstruct a 512³-voxel volume from 360 512²-pixel projections. This performance is 18% faster than the prior method. Some detailed analyses are also presented to understand how effectively the acceleration techniques increase the reconstruction performance of a naive method.