Distributed Multi-GPU Accelerated Hybrid Parallel Rendering for Massively Parallel Environment

For lack of local graphics hardware, general purpose massively parallel computer can hardly be used to quickly visualize the large-scale locally generated datasets with hardware-accelerated volume rendering technique. In this paper, we presented a hybrid parallel rendering scheme take advantage of the background massively parallel computer and front-end distributed graphics workstations. The streams of OpenGL rendering commands are issued from the multi-processors of parallel machine, and the GPU-based volume rendering is executed on the remote multi-workstations. The entropy based remote data reduction is used to extract features from the massive datasets, which contributed to the reducing of the data transfer delays during remote rendering. The image compositing is performed on parallel computer to get better communication performance. Hybrid rendering performance can be further improved by overlapping rendering, compositing and display process with parallel pipeline technology. Experimental results show that our hybrid parallel rendering scheme proposed in this paper can reach interactive rates while rendering 17 GB volume dataset in the massively parallel environment.