During the quantization of the hologram values part of the information is lost, which introduces quantization noise in the reconstruction. To enhance the hologram quality, we developed in a previous work a view-dependent error diffusion algorithm which rejects the quantization noise in specific views. However, this algorithm is slow which prevents its use in real-time applications. In this paper, we present a GPU implementation of this technique which uses the inherent parallelism of the error diffusion algorithm. Moreover, we introduce a technique to select diffusion weights, which allows a trade-off between computation time and reconstruction quality. Experimental results demonstrate the ability of our approach to quickly generate high-quality holograms.