Pipelined Sort-last Rendering: Scalability, Performance and Beyond

We present in this paper a theoretical and practical performance analysis of pipelined sort-last rendering for both polygonal and volume rendering. Theoretical peak performance and scalability are studied, exhibiting maximum attainable framerates of 19 fps (volume rendering with back-to-front alpha blending) and 11 fps (polygonal rendering with Z-buffer compositing) for a 1280◊1024 display on a Gigabit Ethernet cluster. We show that our implementation of pipelined sort-last rendering on a 17-node PC cluster can nearly sustain these theoretical figures. We finally propose possible enhancements that would allow to go beyond the maximum theoretical limits. This paper clearly shows the potential of pipelined sort-last rendering for real-time visualization of very large models on standard PC clusters.