Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!uunet!mfci!rodman
From: rodman@mfci.UUCP (Paul Rodman)
Newsgroups: comp.arch
Subject: Re: Evans and Sutherland quits the superbusiness
Keywords: Killer Micros
Message-ID: <1128@m3.mfci.UUCP>
Date: 20 Nov 89 15:13:52 GMT
References: <27611@dhw68k.cts.com> <38966@lll-winken.LLNL.GOV> <1989Nov19.155445.27287@hellgate.utah.edu> <38980@lll-winken.LLNL.GOV>
Sender: rodman@mfci.UUCP
Reply-To: rodman@mfci.UUCP (Paul Rodman)
Distribution: usa
Organization: Multiflow Computer Inc., Branford Ct. 06405
Lines: 60

In article <38980@lll-winken.LLNL.GOV> brooks@maddog.llnl.gov (Eugene
Brooks) writes: ....[ deleted soapboxing about Killer Micros]....

Ok, ok, so I'm wasting my time, but I just can't let Mr. Brooks
continue his deluge of propaganda without throwing in my opinions....

Let's ignore the fact that the Killer Micros usually don't have a
decent memory system or I/O system. [Not to mention that their
compilers and O/S are often not very robust...:-)]

There IS a grain of truth in what Mr Brooks says, in that I do belive
that there has been, and will continue to be both a factory cost and
performance compression from the lowest to the highest performance
computers.

Today's dense CMOS and ECL ASICS appear from PCs to workstations to
supercomputers, as do pretty dense packaging techniques.  So it IS
getting harder to figure out how to apply current technology to build
a supercomputer, i.e. how do I use more hardware to build a faster
uniprocessor machine?

<Commercial ON.>

One technology exists that CAN use more hardware for more performance:
the VLIW machine. Furthermore, such a machine uses replicated
functional units, lots of SRAM, and minimal control, all of which
contribute to ease of the design cycle. The ease of use of the VLIW
system [compile-and-go] is important for the Non-Brookses of the
world, belive me.

I have seen the state of typical software efforts in the good 'ol USA
and I feel that it will be a long time before efforts such as that of
Mr Brooks are the norm. [100's, 1000's of micros used for
time-to-solution]. God knows enough folks are working on, and have
worked on, the problem!  As such multiprocessing techiniques improve,
they WILL be used commercially, but mostly with more modest numbers of
VLIW "supercomputers".

I DO agree with Mr. Brooks that single cpus built with 100's of
different ASIC designs, dozens of boards, and mediocre performance, are
dead meat in the not-so-long run [e.g. Cyber 2000]. But Mr. Brooks
errs in thinking that single-chip cpus are the be-all, end-all of cpu
design. 

<Commercial Off.>

Speaking as a engineer that does NOT work on Killer Micros, I just
don't seem to feel as depressed about the future of high-$$$ cpu
design as he thinks I should.... :-) Mr Brooks' attitude reflects his
rigid thinking, and his underestimation of change in areas other than
the one he has focused on.  His refrain reminds me of all the wags 10
years ago that claimed "The Attack of the Killer CMOS" would remove
LSI ECL from use in computer design.  Instead almost all high-end cpus
sold today are ECL.


Paul K. Rodman / KA1ZA /   rodman@multiflow.com
Multiflow Computer, Inc.   Tel. 203 488 6090 x 236
Branford, Ct. 06405