The ramp preserving 2D nonlinear complex-diffusion filter introduced by Gilboa et al. (2004) was extended to 3D (Maduro et al., 2012). We propose a graphical processing unit implementation of the 3D filter for an overall faster processing in order to be used in a clinical setting. We perform a search for the best diffusion parameters (the number of iterations and spread of the diffusivity) for the 2D and 3D filters and compare their results resorting to synthetic spectral-domain optical coherence tomography volumetric data and several quantitative metrics. Execution time improvement of our implementation versus a single-core approach is also presented, showing that it allows for a full 3D volume to be processed under 7.5 seconds.