Volume Visualization of High Resolution Data Using PC-Clusters

There is a need to visualize multi-billion voxel data sets. Hardware accelerated rendering currently available on PCs provides a cost effective and scalable way to render these very large data sets. Very large voxel data can be rendered on systems with limited texture and system memory piecemeal by decomposing the data into a set of tiled sub-domains, which are sequentially read from disk. Hence, voxel resolution is only limited by disk space. To provide a trade off between image quality and overall rendering speed, and particularly for interactive modes of display, it is useful to generate (prior to rendering) a sequence of reduced resolution representations organized in a tree hierarchy. Such a “Block-Tree Hierarchy” may be thought of as a 3D mipmap. For rendering on clusters of PCs, it is useful to encapsulate the rendering application as a component object module, which may be invoked on remote hosts. Rendering parallelism is possible both across frames in a movie and within and individual frame by decomposing the rendering of an image into a tiling of sub-images, which divides the work of both casting voxels and filling pixels. Implementations of these ideas, and their use in visualizing billion voxel volumes on PCs, are described here.