From: pepke@SCRI1.SCRI.FSU.EDU (Eric Pepke)
Subject: Re: Connection Machine
Date: 22 Mar 91 19:40:36 GMT
Organization: Florida State University, but I don't speak for them



In article <18646@milton.u.washington.edu> frerichs@ux1.cso.uiuc.edu 
(David J Frerichs) writes:

> For graphics, the CM2 has a special frame buffer I/O module that supports
> 40Mbyte/sec transfer rates.  That translates to 13 24bit color megapel 
> pictures per second... quite enough for continuous movement. (Although, 
> that is the peak rate, nominal operations would cut it down a couple 
> frames per second.)

I have to interject here.  There's a CM2 almost right across from my 
office.  We have the framebuffer, and I've played with it a bit.  John 
Murray, who works with me, has played with it even more.  It is NOT a 
machine I would pick for virtual reality development.  In practice, the 
actual throughput to the frame buffer in 24-bit mode is quite poor and 
irregular, especially under timesharing.  (8-bit is significantly better) 
The arrangement of the framebuffer makes it easy to assing a processor to 
a group of nearby pixels.  This makes it quite good for 2-D colormapped 
displays of 2-D problems.  It's also pretty good for fast first order 
raytracing of SIMPLE scenes.

For more complex geometry, it really doesn't do the trick.  There are
some parallel line-drawing routines which are glacial.  It might be
possible to do some pretty good geometry routines using a strategy 
similar to the Pixel Planes project, but then the total polygon 
throughput will be limited by the clock speed of the machine.  Our 
obselete IRIS 4D-240 GTX blows it away for doing geometry.

Furthermore, the frame buffer board does not do RS-170A.  It comes fairly 
close, but not close enough to sync up, say, a Sony laser videodisc 
recorder.  Our cheap-o Lenco encoder will make a semblance of an NTSC 
signal out of it, with some peculiarities.  It might or might not drive an 
eyephone, depending on how picky it is, and it might or might not get the 
colors correctly.  If you already have a scan converter, you probably 
don't care.

That having been said, I love the Connection Machine.  Its architecture is 
great for certain problems, and you can get gigaflops out of it on real 
live research codes.  However, it is far from optimized for graphics.

Eric Pepke                                    INTERNET: pepke@gw.scri.fsu.edu
Supercomputer Computations Research Institute MFENET:   pepke@fsu
Florida State University                      SPAN:     scri::pepke
Tallahassee, FL 32306-4052                    BITNET:   pepke@fsu

Disclaimer: My employers seldom even LISTEN to my opinions.
Meta-disclaimer: Any society that needs disclaimers has too many lawyers.