NetNews Usenet Archive 1992 #26/NN_1992_26.iso/spool/comp/arch/10394

Internet Message Format | 1992-11-04 | 5.5 KB

Path: sparky!uunet!charon.amdahl.com!pacbell.com!decwrl!sun-barr!ames!haven.umd.edu!darwin.sura.net!europa.asd.contel.com!paladin.american.edu!news.univie.ac.at!hp4at!mcsun!sunic!psinntp!psinntp!parlan!richardw
From: [e-mail] (Richard Walter)
Newsgroups: comp.arch
Subject: Re: QED's low power R4000 @ Microprocessor Forum
Message-ID: <[e-mail]>
Date: Tue, 03 Nov 92 18:04:11 PST
References: <[e-mail]>
Organization: Parallan Computer, Mountain View CA
Lines: 97

[e-mail] (Andy Glew) writes:

>
> This is an area that I don't have too much background in, so mind if I
> ask a naive question?:
>
> Which is preferable in this Brave New World of really small portables?
>
> (1) A small low power chip that requires an external cache?
>
> (2) A slighter larger, slightly more power hungry, chip that does not
> require an external cache?
>
> I.e. how do people building such systems want the tradeoff between
> number of devices and power consumption to be made?

Well, having worked at a company which makes laptops/notebooks/pen-based
machines (although I don't work there anymore), I'll give my view of the
state of affairs...

What the question really comes down to is: Do people making small
computers want large chips with lots of integration and lots of features,
(and a larger power consumption) or smaller chips, needing external help
for features and smaller power consumption?

Most people would say (and it seems that most chip companies are thinking)
that the first option, large, feature-ful chips, are preferable.  Their
reasoning being that, Yes, these chips do have a higher power consumption
than the smaller one, BUT the consumption of a single integrated solution
will be smaller that the collective consumption of many smaller chips to
do the same functions.

However, I feel that the smaller, less power chips, are what the small
computer designers want.  There are several reasons for this:

Differentiation:  If company X puts out a 1 or 2 chip set which solves
"all of the problems for you." and PC makers A, B, & C all use it.  Then,
A, B & C will have all of the same features, same drawbacks, and same
performance.  This is not what the PC maker's marketing department wants
at all.  They want to say how much better they are then the other guy's
machine.  By using common hardware, the only differentiation is software,
which can do some, but not all.

PCB Real Estate:  Look at almost any desktop/workstation PC board and you
will see nice rows and columns of DIPS (or perhaps surface mount devices).
and the PC board itself will be roughly square.  Now, look at the MLB in
a typical notebook (or handheld) and you will see PC boards with odd
cut out areas.  Add to that the height restrictions due to the cases,
peripherals (PCMCIA cards are thin, true; but they are also wide and long),
screws (you do not want the machine to fall apart when it is dropped), and
even itself (some compaq machines, for example, FOLD the MLB across itself
with flex circuits in 3 places!) and you don't have a regular structure
in which to place your chips.  Now, just because there is 20 square inches
of board space does NOT mean that you can put down 20 inch square chips.
The reason is because finding a contiguous space of 1 inch on a side is
not easy.  However, fitting physically smaller chips is easier because
they can fit between screws and don't have to be placed facing the same
direction.

Functionality:  In an ideal world, this wouldn't be a consideration, but
hey, we do have problems.  An example here might explain this better:
In the Intel 386SL chip set, the I/O chip has 2 RS-232 serial ports
integrated into it.  However, they don't entirely work correctly.  (I
don't mean to Intel-bash here, but it is true).  And, they don't want to
fix it because the error is (according to them) not that bad.  (Except
that LapLink 3 and Windows both won't work with them.)  As a result,
people who wanted correctly working serial ports had to put down an
external serial controller anyway, and the space taken up by the SL chip
was wasted.  If they had not integrated the entire I/O devices together,
then just the serial chip could have been swapped.

Pin Limitations:  Most chips these days are Pin limited, not Silicon
limited.  Which means that technologically, adding more functions to the
chip is easy.  However, to do so adds more pins, and adding more pins
makes the physical size of the chip grow larger at a faster rate than
the functionality grows.

General comments:  The original question asked about integrating/excluding
a cache on a processor.  My personal preference is to say external.  But
the question also implies something subtle:  He wants high performance
out of a small computer.  Now there are problems for which high performance
is required.  And for those, you design fast, complex processors.
Anyone could build a small machine with a chip such as DEC's Alpha, and
it would be one hell screamingly fast notebook....for about 30 seconds,
and then your battery would die a horrible death, begging for forgiveness
for all of it's earthly sins.

For the chip makers, higher integration gives the chance for showing off
and one-upmanship among their peers.  And I don't see that trend ending,
however, I feel that for truly flexible and unique architectures, smaller
mix-and-match type parts are needed.  This is especially true in the
handheld and sub-notebook markets where the available space disappears
very quickly.

-Sincerely,
Richard Walter
[e-mail]