[comp.graphics] best real-time 3D graphics

don@andante.UUCP (Don Mitchell) (03/06/90)

SGI Iris and AT&T Pixel Machine are the two high-end real-time 3D
Graphics machines that I know well.  We have one of each.

The Iris uses fast CPU(s) in combination with special-purpose polygon
rendering hardware.  The PXM is based on high-speed signal processing
chips.  A PXM advantage is that you can program everything inside it
(in C).  For example, John Amanatides and I wrote a ray tracer for it.
Others have done real-time textures, bump maps, environment maps, etc.

kyriazis@iear.arts.rpi.edu (George Kyriazis) (03/06/90)

In article <31232@andante.UUCP> don@andante.UUCP (Don Mitchell) writes:
>SGI Iris and AT&T Pixel Machine are the two high-end real-time 3D
>Graphics machines that I know well.  We have one of each.
>
>The Iris uses fast CPU(s) in combination with special-purpose polygon
>rendering hardware.  The PXM is based on high-speed signal processing
>chips.  A PXM advantage is that you can program everything inside it
>(in C).  For example, John Amanatides and I wrote a ray tracer for it.
>Others have done real-time textures, bump maps, environment maps, etc.

I agree that the Iris is a very nice graphics engine that is suited for
real-time 3d graphics.  On the contrary, I don't think that the PXM
is suited for real-time stuff.  It might have fully programmable CPU's, but the
pixels <--> CPUs mapping is such that you have to use the pixel nodes just
for rendering.  In other words you waste 10Mflops * 64 = 640Mflops to
draw pixels.  The transformations have to be done with the pipe nodes.
If you are talking about rendering polygonal meshes, your 'fully programmable'
pipe nodes do not help at all.  The pipeline bottleneck still exists, but
now it's more pronounced, since the pipeline processing elements are not
as fast as the SGI's geometry engines, for example.  
I don't disagree though that the PXM is a great ray-tracing machine though.
(I wrote a ray-tracer for it too!)



----------------------------------------------------------------------
  George Kyriazis                 kyriazis@turing.cs.rpi.edu
				  kyriazis@rdrc.rpi.edu
 				  kyriazis@iear.arts.rpi.edu