The interactive visualization of large astrophysical datasets is a formidable challenge. Cosmological simulations performed on current terascale facilities typically generate datasets up to 100 gigabytes in size. In order to understand the complex three-dimensional structure within these simulations, interactive visualization with rapid rotation and zooming is required, but handling this much data is well beyond the capability of even current high-end graphics workstations. Rendering can be performed on a parallel computer, but the user typically is using such facilities remotely. Both the latency and bandwidth between the user’s workstation and the parallel machine preclude the possibility of interactive rendering.
Techniques from the computer graphics community were used to overcome the latency and bandwidth issues. This work explores the use of 3D voxel grids to represent particle data. The voxels are rendered as 3D textures on the parallel server, and communicated to the client where they are stored as 3D textures directly on the client’s graphics card. Real-time rotation can then be performed by the graphics hardware.
These techniques have been implemented within a framework that is currently being used in research on galaxy formation and evolution, and is sufficiently extensible that it can be used in a variety of applications, from planet formation simulations to catalogs of galaxies from large-scale structure surveys. Data from all sources can be rendered in real-time 3D.
This work was done by Thomas Quinn of the University of Washington; Orion Lawlor of the University of Alaska, Fairbanks; and Laxmikant Kale of the University of Illinois, Urbana- Champaign for Goddard Space Flight Center. GSC-16812-1