2400T vs. 4D performance

[Steve Dempsey]

When porting line drawing applications from 3000/2400T systems to the GT
one needs to take SGI's advice in the document entitled "Tuning Graphics
Code for your IRIS-4D" very seriously.

I recently attempted a port from a 2400T to a 4D/240GTX of an application that
makes extensive use of SGI's display lists and the "old style" line drawing
routines.  Since I was not using the high performance GT drawing functions
I didn't expect the ~5x performance improvement one might anticipate
from the raw hardware specs (~100K vs. ~500K vectors/sec).  However, I was
quite dismayed to discover that the GTX version generated only HALF as
many frames per second as the 2400T version.

Profiling revealed that the display list interpreter was consuming 13% of
the CPU time on the GTX vs. only 0.1% on the 2400T.  The use of "gr_osview -a"
showed that the 4D processor my application was running in was idle 75%  of 
the time due to delays in waiting for the graphics pipe to empty (a side effect
of using the "old style" drawing routines, I presume).

The porting guide mentioned above indicates that line drawing will occur
at rates up to 10x faster if one uses the "high performance" line drawing
functions, which don't work in display lists.  Read another way, this means that
the "old style" functions in display lists are ~10x slower.  So given hardware
that is ~5x faster in combination with software that is ~10x slower, I can see
why in this particular case the GTX is only half as fast as the 2400T.

Question:
	Do the non-GT and Personal IRIS 4D workstations also suffer from poor
	display list performance?
-------------------------------------------------------------------------------
Steve Dempsey						          (619) 534-0208
Dept. of Chemistry Computer Facility, B-014	   INTERNET:   sdempsey at ucsd.edu
University of Calif. at San Diego		     BITNET:   sdempsey at ucsd
La Jolla, CA 92093				       UUCP:   ucsd!sdempsey