Vis5d Clear Air Turbulence Benchmark


The Vis5d Clear Air Turbulence benchmark measures graphics performance on triangle and quadrilateral mesh primitives of the Vis5d application running on the NCAR Clear Air Turbulence dataset. Vis5d is an Earth Sciences visualization utility available from the University of Wisconsin. The version of Vis5d used in these benchmarks is based on Vis5d release 5.2. The only relevant modifications made to the official 5.2 release is the addition of timers needed to support performance profiling. The Clear Air Turbulence (CAT) dataset are multi-variate, multi-timestep computer-modeled weather data.

Vis5d is capable of using a number of visualization techniques for representing data. This benchmark is constrained to isosurfaces, represented as triangle meshes; direct volume rendering, represented as alpha-blended quadrilateral meshes; and contour plains, again represented as texture-mapped quadrilateral meshes.

All primitive rendering is performed in immediate mode - display lists are not used.
 

Isosurfaces

Isosurfaces are rendered as opaque, lit, triangle meshes. Verticies consist of a 2-byte per coordinate position vector (6 bytes total) and a 1-byte per coordinate normal vector (3 bytes total). Z-buffering is enabled. Anti-aliasing is disabled.
 

Contour Slices

Contour slices are rendered as opaque, unlit, textured, quadrilateral meshes.  Vertices consist of a 2-byte per coordinate position vector (6 bytes total) and a 4-byte per coordinate texture vector (4 bytes total). Z-buffering is enabled. Anti-aliasing is disabled.
 

Volume Rendering

Volume rendering is performed using stacks of as alpha-blended, unlit, quadrilateral meshes.  Vertices consist of a 2-byte per coordinate position and a 1-byte per coordinate color/alpha vector. Z-buffering is disabled and anti-aliasing are disabled.
 

Data

The CAT dataset used in this benchmark consists of 11 timesteps  at 240x240x72 resolution, containing three scalar variables: NSTRO, VRMS, and VS. The scalar values are byte-scaled; the total dataset size is approximately 11 * 240 * 240 * 72 * 3 = 136MB's.
 
 

Experiments

The benchmark is run for each of the three visualization techniques (isosurface, chslice, volume), each at three different spatial resolutions (increasing fill requirements while maintaining constant T&L requirements).

For each experiment, the primitive rate in Megaprimtives/sec is measured. The total frame rate in frames per second is also computed.

This page maintained by John Clyne (clyne@ncar.ucar.edu)

$Date: 2000/04/17 05:58:39 $, $Revision: 1.1 $