[MUD-Dev] Re: Wild idea.. :)

[Adam Wiggins]

> On Wed, 2 Sep 1998, Marc Hernandez wrote:
> > 	Which happily (though assembly isnt bad) is not much anymore.  Now
> > it seems to be making sure you only send the rasterizer x many triangles
> > (where x is somewhere between 3000 and 8000 per frame in batches of 30-50)
> > and sort the triangles based on what texture they are.

Just a quick technical note: my work project is using a fairly
well-optimized (good culling, plenty of assembly for the matrix
math, etc) custom renderer based on the Glide rasterizing library.  On our
prototype Katmai 500mhz (aka Pentium III) with a Voodoo2, running in Linux
console mode (ie, practically zero overhead from the OS) it can only put
through 3000 or so polygons and maintain full framerate (60fps).  Even if
your standards are lower (say, 30fps) you're still only talking ~5000
polys in a scene.  I can't speak for other platforms at the moment, JC
could probably fill you in on SGI throughputs.

Hardware is definitely getting better, but 3D still has a long way to go
before we can stop worrying about complexity.  For a more realistic
example, the K6-233 with a Voodoo1 and (Mesa) OpenGL does around 1000
polys in the scene at 30fps.  Obviously this number can vary alot
depending on the average size of the polys, your system's bus speed,
overhead from other game stuff (AI, input polling, sound mixing), screen
resolution, and more.  But I'd say that expecting to put 8000 polys into a
scene and have it run nicely on your run of the mill computer is a little
out of reach for another year or two.

Adam