Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cs.utexas.edu!sun-barr!decwrl!jumbo!ehs
From: ehs@jumbo.dec.com (Ed Satterthwaite)
Newsgroups: comp.sys.mac.hardware
Subject: Re: 68030 fabrication technology (was Re: Why 68000?)
Summary: CPU technology, balanced systems
Message-ID: <14561@jumbo.dec.com>
Date: 18 Feb 90 21:59:43 GMT
References: <1990Feb11.154304.19943@smsc.sony.com> <3919@hub.UUCP> <1990Feb16.195520.6853@intercon.com>
Organization: DEC Systems Research Center, Palo Alto
Lines: 53

In article <1990Feb16.195520.6853@intercon.com>, amanda@mermaid.intercon.com (Amanda Walker) writes:
> 
> They might be able to drop the power my going to all high density CMOS,
> but the die size would increase, which could make for a yield problem.  Maybe
> Motorola is waiting to get higher yields for smaller design rules before they
> go to all CMOS, or maybe it just doesn't make much difference.  We're
> starting to get out of my league here; what we need now is a VLSI person
> from Motorola...

Well, in no way do I speak for Motorola, and I am a board designer, not a
"VLSI person," but ...

Most "CMOS" microprocessors are in fact fabricated from a mixture of CMOS
and NMOS technology.  Roughly speaking, a CMOS circuit requires more
transistors than its NMOS equivalent, so that power consumption can be
traded for die size within certain limits.  Also, the power consumed by a
CMOS circuit increases as the switching speed (clock frequency) increases.
Thus the power advantage decreases as speed increases.  Also, one can get
hints about the dominant technology by looking to see if power dissipation
is specified as a function of frequency.  Here are some numbers pulled
from Motorola's 680x0 data books.  All are maxima and are functions of
temperature, packaging option, etc., but the ratios should be about right:

  uP            Pd (watts)
68000, 68010   1.5-1.7
68hc000        0.13     8 Mhz
               0.19     12.5 MHz
               0.26     16.7 MHz
68020          2.0
68030          2.6

Comments on the original query:

For any given level of volume discount, the 68000 and 68hc000 are a *lot*
cheaper than the 68030.  But to take anything like full advantage of a
68030, including its 32-bit bus and higher clock speeds, a designer must
pay a lot of other costs: twice as many bus buffers and transceivers,
twice as many packages in minimum memory configurations (or expansion
increments), perhaps a second-level cache built from high-speed SRAM, more
expensive PC board technology, more cooling, etc.  This gets expensive
pretty quickly.  IMHO, many of the existing 68030 (and 80386/80486) PC
systems are catering to processor envy; the money would be better spent on
upgrading the memory and i/o sytems than on the very latest, fastest CPU.
Also, as I understand things, hw support for virtual memory will not be
required for System 7 and, again IMHO, its use is a very bad idea on a Mac
without a big, fast disk (maybe even then :>).

Finally, in a related posting, Amanda Walker asks about the 68012.  I
believe that a recent issue of Motorola's Update announced the last chance
for lifetime buys of this part.  It's gone from the current price book.

Ed Satterthwaite
ehs@src.dec.com  or {...}!decwrl!ehs