Back to Research

Large Datasets

Interactive Rendering of Arbitrarily Large Datasets is a fundamental problem in computer graphics and scientific visualization and a critical capability for many real applications. Interactive visualization of large datasets poses substantial challenges.

The visualization pipeline may be broken down into four major stages: retrieval from storage, processing in main memory, rendering in the Graphics Processing Unit (GPU), and display on the screen. The performance of each of these stages is limited by several potential bottlenecks (e.g., disk or network bandwidth, main memory size, GPU triangle throughput, and screen resolution). iRun uses out-of-core data management and speculative visibility prefetching to maintain a working-set of the geometry in memory. Our rendering approach uses GPU-assisted volume rendering with a dynamic set of tetrahedra and uses an out-ofcore LOD traversal. Finally, our system is implemented in VTK and allows distributed rendering for high-resolution displays. Using a single commodity PC, our system can render datasets consisting of 14 million tetrahedra while maintaining interactive frame rates.