Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!usc!zaphod.mps.ohio-state.edu!caen!gilgalad
From: gilgalad@caen.engin.umich.edu (Ralph Seguin)
Newsgroups: comp.sys.amiga
Subject: Re: Sorry Guys, There is NO WAY!
Message-ID: <1990Dec5.154228.20653@engin.umich.edu>
Date: 5 Dec 90 15:42:28 GMT
References: <9012032130.AA114840@vttcf.cc.vt.edu> <21968@well.sf.ca.us>
Sender: news@engin.umich.edu (CAEN Netnews)
Organization: University of Michigan Engineering, Ann Arbor
Lines: 36

In article <21968@well.sf.ca.us> yoo@well.sf.ca.us (Young-Kyu Yoo) writes:
>fmcphers@VTTCF.CC.VT.EDU (Frank McPherson) writes:
>>	You say that the 3000ux will "crank out 5 mips", then say the 040 NeXT
>>will "crank out 15-20".  Figures courtesy of Motorola, you say.  I have no 
>>reason to doubt you, so I won't say anything about that, except for being 
>>interested in how such an answer was arrived at.

These were probably dhrystone 2.1 derived results.  A bad indicator of
speed at best.

>>	I do, however, take exception to the assertions concerning the 
>>MFLOPS of both machines.  "nearly 3" and "will be lucky to approach 1" isn't
>>precise enough for this arguement.  They suggest that you don't know for sure,
>>you're just pulling out some numbers which sound nice.  Don't post something

Nope.  This is pretty much accurate.  Running things like ray-traces, FFTs,
etc, as well as some of my own code, as well as the usual LINPACKs, Livermore
Loops, etc indicates that the 040 ranges from 2.5 MFLOPs to 3.9 MFLOPs.
The 882 at 25 MHz, on the other hand, gives figures from 0.075 MFLOPs to
.453 MFLOPs.  Even in primitive multiply/divide loops, I've been unable to
beat .5 MFLOPs on an 882.  All of this depends on the instruction mix that is
used.  With lots of transcendental stuff, the perf on both chips goes way
down.  With just arithmetic, the results are higher.

>I assume the math coprocessor on the Amiga 3000ux is Motorola's 68882.
>68882 performs at less than 1 Mip.  If you have any info to contradict this,

I'll guarantee you that the 882 is less than one MIP :)  It is also less
than .5 MFLOPs at 25 MHz.  I haven't had the opportunity to benchmark a
50 MHz 882, but maybe later on today.

			See ya, Ralph

Ralph Seguin			gilgalad@dip.eecs.umich.edu
536 South Forest Apt. #915	gilgalad@caen.engin.umich.edu
Ann Arbor, MI 48104		(313) 662-4805