[comp.graphics] What is more important????

) (03/16/89)

In trying to determine the best route in which to upgrade my Amiga 2000 to 
reduce the time needed to do ray tracing, I need to know if it would be 
more important to obtain an upgrade board with the following:

68020 running at the normal 7.14Mhz clock speed, with a 68881 running at 
16Mhz(full 32 bit path used), the reduced speed on the 68020 ensures 
compatability with all software. Expected speed increase for normal prgs. is
62%(ones that do not use the math chip). Price $500.00 US(includes the 68020
and 68881).

Or I could get a full 16Mhz board with a 68020 running at full speed, with 
a 16Mhz 68881. Price about $1800.00 US(would have to also pay a fortune for
the 32-bit ram).

My big question is this: "Under normal ray tracing, would the prg. make more
use of the 68881 or the 68020(which is more important)? Of course the program
itself only uses the 68020, but how much math is really being done???



later
Barry Comer   870646c@aucs.UUCP

greg@bilbo (Greg Wageman) (03/21/89)

In article <1663@aucs.UUCP> 870646c@aucs.UUCP (The Buck Stops Here!) writes:
>In trying to determine the best route in which to upgrade my Amiga 2000 to 
>reduce the time needed to do ray tracing, I need to know if it would be 
>more important to obtain an upgrade board with the following:
>
>My big question is this: "Under normal ray tracing, would the prg. make more
>use of the 68881 or the 68020(which is more important)? Of course the program
>itself only uses the 68020, but how much math is really being done???

Ray-tracing is one of the most floating-point-math-intensive
applications I know of on personal computers.  Using the fastest math
coprocessor available for your machine will be a big performance win.
Of course, running your main processor at a higher speed will help
too, since it is still responsible for executing the overall program.
But in this application the math coprocessor is heavily used,
performing literally hundreds of millions of floating-point operations
for a typical scene, say 640 by 400 pixels, with reflections and/or
refractions.


Longish .signature follows.  Skip now, or don't complain!

Greg Wageman			ARPA:  greg@sj.ate.slb.com
Schlumberger Technologies	BIX:   gwage
1601 Technology Drive		CIS:   74016,352
San Jose, CA 95110-1397		UUCP: ...!uunet!sjsca4!greg
(408) 437-5198
------------------
There's nothing I hate more than a Usenet posting which took three
seconds to compose and three minutes to type, glibly dismissing three
years (or three decades) of an author's work in three lines.
------------------
Opinions expressed herein are solely the responsibility of the author.
(And the author wouldn't have it any other way.)

cdouty@jarthur.Claremont.EDU (Christopher Douty) (03/26/89)

In article <1663@aucs.UUCP> 870646c@aucs.UUCP (The Buck Stops Here!) writes:
>In trying to determine the best route in which to upgrade my Amiga 2000 to 
>reduce the time needed to do ray tracing, I need to know if it would be 
>more important to obtain an upgrade board with the following:
>
>68020 running at the normal 7.14Mhz clock speed, with a 68881 running at 
>16Mhz(full 32 bit path used), the reduced speed on the 68020 ensures 
                                                        ^^^^^
>compatability with all software. Expected speed increase for normal prgs. is
>62%(ones that do not use the math chip). Price $500.00 US(includes the 68020
>and 68881).
>
>My big question is this: "Under normal ray tracing, would the prg. make more
>use of the 68881 or the 68020(which is more important)? Of course the program
>itself only uses the 68020, but how much math is really being done???

>Barry Comer   870646c@aucs.UUCP


First off, the speed of the 68020 has no bearing on compatability.  If the
program was poorly written so that it will GURU a '020 machine, it will
bring the walls tumblin' down no matter what the clock speed.

Apparently in programs which were not specifically compiled with in-lin FPU
code the speed of the CPU (68020) makes a bigger difference than the FPU.
Normal Sculpt-3d (not the '881 version) renderings are about twice to four
times as fast with a 16Mhz 68020.  

To really make a difference in your ray-trace times, you should get both the
CPU and FPU going as quick as you can and use programs compiled specifically
for that pair.  You might want to check out the A2620 board from Commodore.
It will plug right into your 2000 with no fuss, and it includes 2 megs
32-bit RAM for about the same price as a RAM-less CSA or Ronin board.
Another possible choice would be Great Valley Products new '030 board.  It
probably costs a mint but is disgustingly fast, has 32-bit RAM, AND a
built-in SCSI controller in addition to a 25MHz (?) 68030/68882 combo.

I hope that this random rambling helps.  I know that a 68020 makes
everything on the machine much snappier and can vouch for the A2620.

Christopher Douty
cdouty@jarthur.claremont.edu

with STANDARD_DISCLAIMER;  use STANDARD_DISCLAIMER;
with SILLY_QUOTE;  use SILLY_QUOTE;