Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!eecae!netnews.upenn.edu!rutgers!bellcore!texbell!merch!cpe!hal6000!trsvax!johnm
From: johnm@trsvax.UUCP
Newsgroups: comp.graphics
Subject: Re: Ray Trace for IBM
Message-ID: <194300054@trsvax>
Date: 15 Feb 89 14:43:00 GMT
References: <986@novavax.UUCP>
Lines: 51
Nf-ID: #R:novavax.UUCP:986:trsvax:194300054:000:2765
Nf-From: trsvax.UUCP!johnm    Feb 15 08:43:00 1989


>>Doesn;t  ray  tracing take up huge amounts of cpu time?  Does anyone
>>have examples  of  cpu time/ frame on various machines?
>
>Well, last time I checked my raytracer on a 20Mghz 68020 and 68881,
>I managed to trace a scene with 8192 reflective spheres in ~2.5 hrs.
>Unfortunately for the PC, segmented memory confines you to small numbers of 
>objects, or use large model (or huge, or superhuge, or whatever) which
>slows all pointer dereferencing.  I've never really played to much with 
>PCs, so I don't have any hard figures for them.

As Mark says, the available ray tracers for the PC have all been compiled 
using the large model and that makes a significant impact on the speed of the
ray tracing.  A very good example is using the QRT v1.4 ray tracer which I have
timed on both a 20Mhz 386 machine (the Tandy 4000LX) and on an 8Mhz 68000
machine (Amiga 2000) at home.  Normally a 68000 is comparable in speed to a 
286 machine in terms of overall speed but when running this particular program 
on both machines (without floating point chips in either) the Amiga will 
finish in about the same amount of time.  Large numbers of floating point
operations combine with the segmented architecture to kill the speed of the
machine.

>People in general overestimate the amount of CPU time that raytracing takes.
>Improvements in hierarchy methods have made it quite feasible to use
>raytracing.  Certainly the next generation of workstations will help.
>Combined with the generality and (relative) simplicity of the method, 
>raytracing is in many ways a superior algorithm to many scanline algorithms.

I agree with this within limits.  There are still some things that only a
scanline method is practical for, like generating animation.  A good scanline
renderer on a fairly fast machine may have a new frame ready every 3-4 minutes
depending on the complexity of the scene.  No ray tracer that I have scene
could possibly do that (on the same hardware).

>>I would think  several hundred hours per frame on an ibm might not
>>be too far off...

DBW Render for the PC will probably live up to your expectations :-).  Look
over in comp.binaries.ibm.pc for a copy of it.

However, QRT v1.4 which can be gotten on Compu$erve in the Turbo C 2.0 data
library of the Borland section (GO BPROGB) will be able to do ray tracings in
much more reasonable times like 5-6 hours depending upon what hardware you
have.  If you can add a floating point accelerator to your equipment it is
possible to get QRT to spit frames out in 15-20 minutes (Tandy 4000LX & 80387).

>People who have been working on my raytracer have reported times on the order 
>of a day for moderately complex images rendered on the PC.  It's all in the
>tuning and optimization....

John Munsch