Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!ames!lll-winken!uunet!pdn!rnms1!alan
From: alan@rnms1.paradyne.com (0000-Alan Lovejoy(0000))
Newsgroups: comp.sys.next
Subject: Re: Selling through Businessland
Message-ID: <5893@pdn.paradyne.com>
Date: 1 Apr 89 23:00:24 GMT
References: <1554@neoucom.UUCP> <121@dg.dg.com> <689@garcon.cso.uiuc.edu>
Sender: news@pdn.paradyne.com
Reply-To: alan@rnms1.paradyne.com (0000-Alan Lovejoy)
Organization: AT&T Paradyne, Largo, Florida
Lines: 50

In article <689@garcon.cso.uiuc.edu> dorner@pequod.cso.uiuc.edu (Steve Dorner) writes:
>Don't get me wrong; I'm not saying an 88000 isn't faster than a 68030 on
>real work; I really don't know.  But the obvious calculation of
>
>	    17MIPS / 5MIPS = 3.4 times faster
>
>is wrong, isn't it?  Or is that 17 MIPS figure normalized to VAX MIPS?

It's normalized to benchmark performance in seconds, where the performance
of the latest version of VMS, using the latest compilers, on a VAX 11-780,
is honorarily labeled "1 MIPS" (note that "MIPS" is not plural).  The most
recent VAX 11-780 Dhrystone (v1.1) number I have seen is 1792/second.  

There is about a 15% performance (number/second) decrease in Dhrystone numbers
in version 2 of the benchmark compared to version 1.1.  There is a similar
effect in comparing version 1.0 and version 1.1 of the benchmark.  In other
words, the same machine will execute fewer Dhrystones/second on later (higher
numbered) versions of the benchmark.

RISC CPUs tend to do better on the Dhrystone relative to CISC CPUs than is 
the case with large real-world programs.  Anywhere from 10% to 50% better.  
RISCs with register windows suffer most heavily from this effect (e.g., SPARC).

The 20 Mhz 88k executes 32,000+ Dhrystones/second.  The 25 Mhz 68030 does about 
8000.  That computes to a 3.5x integer performance advantage for the 88k
on Dhrystone v2.1.  The 20 MHz 88k does about 7 MFLOPS SP, 3.5 MFLOPS DP.
This is 10 to 20 times faster than the 68030/68882 chipset.  The MIPS R3000
has very similar performance.  The i860 may be 15% faster at the same clock
speed (integer), but that may be an illusion due to Intel benchmarking
chicanery (in-lining of string functions in Dhrystone).  The i860 FP
performance is 2 to 3 times better than the 88k's.

The 68040 at 25 MHz provides 13.5 MIPS (Motorola's number, no Dhrystones
available.  A number much below 25,000 Dh/s would call the MIPS rating into
question, however), and 2 MFLOPS SP (what benchmarks this is based on has not 
been publicly stated).

By the time the 68040 is available in production quantities (1Q 90), you can
expect the next generation 88k to be in sampling (and available in production
quantities six months later).

DISCLAIMER: These performance characterizations represent averages based on
available data.  It is not unusual for actual relative performance of different
CPU architectures to vary by a factor of 2 or 3 when using different benchmarks
and/or applications!

Alan Lovejoy; alan@pdn; 813-530-2211; AT&T Paradyne: 8550 Ulmerton, Largo, FL.
Disclaimer: I do not speak for AT&T Paradyne.  They do not speak for me. 
__American Investment Deficiency Syndrome => No resistance to foreign invasion.
Motto: If nanomachines will be able to reconstruct you, YOU AREN'T DEAD YET.