Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10    5/3/83; site cornell.UUCP
Path: utzoo!linus!wivax!decvax!harpo!floyd!vax135!cornell!hal
From: hal@cornell.UUCP
Newsgroups: net.arch
Subject: Re: Cray vs ICs - (nf)
Message-ID: <4687@cornell.UUCP>
Date: Thu, 23-Jun-83 17:34:43 EDT
Article-I.D.: cornell.4687
Posted: Thu Jun 23 17:34:43 1983
Date-Received: Sat, 25-Jun-83 22:04:35 EDT
References: <2267@uiucdcs.UUCP>
Organization: Cornell Computer Science
Lines: 36

I recommend the article "The CRAY-1 Computer System" by Richard M. Russell
of Cray Research (CACM vol. 21, no. 1, pp. 63-72, 1978), which explains,
among other things, the reasons for the very simple chip types in the machine.
(The four chip types are 16x4 bipolar register chips [6 ns. cycle], 1024x1
bipolar memory chips (50 ns), and two types of bipolar logic chips.)  A 
paragraph near the end of the article is worth quoting:

"CRAY-1 modules are 6 inches wide.  The distance across the board is about
a nanosecond which is just about the edge time of the electrical signals.
Unless due precautions are taken, when electrical signals run around a board,
standing waves can be induced in the ground plane.  Part of the solution is to
make all signal paths in the machine the same length.  This is done by padding
out paths with foil runs and integrated circuit packages.  All told, between
10 and 20 per cent of the IC packages in the machine are there simply to pad
out a signal line.  The other part of the solution was to use only simple gates
and make sure that both sides of every gate are always terminated.  This means
that there is no dynamic component presented to the power supply.  This is the
principal reason why simple gates are used in the CRAY-1.  If a more complex
integrated circuit package is used, it is impossible to terminate both sides
of every gate.  So all of the CRAY-1's circuits are prefectly balanced.  Five
layer boards have one ground layer, two voltage layers, and then the two logic
layers on the outside.  Twisted pairs which interconnect the modules are
balanced and there are equal and opposite signals on both sides of the pairs.
The final result is that there is just a purely resistive load to the power
supply!"

(Copyright (C) 1977, ACM, copied by permission)


Folks who design supercomputers have strange and wonderous things to worry
about.


Hal Perkins                         uucp:  {decvax|vax135|...}!cornell!hal
Cornell Computer Science            arpa:  hal@cornell
                                  bitnet:  hal@crnlcs