Subject: Test Matrox Millennium OpenGL hardware acceleration for OpenGL.
All tests were done with several screen resolutions that supported
16 bit color, double buffering, and z buffering in hardware.
Test hardware rendering using 16 bit z buffer.
Test software rendering using 32 bit z buffer (not supported by hardware).
Note: tri/sec means triangles per second.
Hardware: 190,000 tri/sec (advertized for some DOS or Windows 3.1 APIs)
My Experiment using Windows NT 3.51, 3D-DDI, driver 2.00.044, Pentium 120:
Hardware: 16,000 tri/sec (independent of number of pixels per triangle)
Software: 16,000 tri/sec (for 12 pixel triangles)
Software: 8,000 tri/sec (for 50 pixel triangles)
Others have reported software being 10 times slower than hardware for
triangles with thousands of pixels.
Note that for triangles of size smaller than 12 pixels, software
rendering (which can be forced by using a 32 bit z buffer)
is actually faster than hardware.
To increase speed (when using hardware):
Reduce the NUMBER of triangles, not their SIZE.
Comment out glCallLists (used to write text to screen).
High screen resolution gives more pixels/triangle.
This penalizes software speed, but not hardware speed.