A major challenge in protein-protein docking is the distinction between near-native and decoy complex predictions. It has been shown that near native solutions are usually located at the bottom of deep and densely populated funnels in the binding energy plot of the complex. Thus exploration, whether the energy plot of the vicinity of a docking solution is 'funnel like', can serve as a validation of such a solution. Generation of such densely sampled plots, however, is a major computational challenge. We have designed an accurate and highly efficient parallel algorithm for generation of such energy plots and implemented it on a server with 4 GPU processors, each with 2880 cores. The algorithm achieved a speedup of about 150 compared to its serial counterpart, while even outperforming it in the achieved results. While the algorithm proved very useful for near native complex hypothesis validation, it still detects many funnels for decoy solutions, especially those with good shape complementarity.