bmfowler@watcgl.waterloo.edu (Barry Fowler) (02/16/89)
I was recently re-reading the SIGGRAPH '88 paper "High Performance
Polygon Rendering" by Kurt Akely and Tom Jermoluk of SGI.
The paper describes the architecture of the graphics subsystem of a
workstation comprised of multiple RISC-based CPU's, presumably the
Power Series. The following is quoted from the summary of the paper:
Benchmark testing of a completed system immediately prior
to publication yielded the following results:
- 101,000 quadrilaterals per second. 100 pixel,
arbitrarily rotate, lighted, Z-buffered.
- 137,000 triangles per second. 50 pixel, arbitrary
strip direction, lighted, Z-buffered.
- 394,000 lines per second. 10 pixel, arbitrarily
directed, depthcued, Z-buffered.
- 210,000 antialiased lines per second. 10 pixel,
arbitrarily directed, Z-buffered.
- 8.3 millisecond full-screen clear. Both color and
Z-buffer banks cleared.
We have a dual-processor GTX, currently being installed. Are these
figures only for the four-processor GTX? The paper does not state
on what kind of machine these benchmarks were run.
I'd like to know how these benchmarks were derived. Were test programs
actual executed on a working machine, or were these numbers computed
based on the characteristics of the graphics system.
If test programs were actually run, I'd like to know if SGI is willing
to distribute the binaries for these benchmarks. Or better yet, how
about the source code? Perhaps such code could teach us a few tricks
on how to make more efficient use of the graphics library.
Would anyone else be interested in this?
- barry
--
"I'm a jazz musician. I'm used to playin' stuff nobody wants to hear."
- Branford Marsalis
bmfowler@watcgl.waterloo.edu