NOTES ON XEROX LISP MACH DEMO

[Rindfleisch]

In response to the message sent  22 Aug 1980 1148-PDT from Cmiller

Following are some notes on the Xerox LISP machine demo last night:

1) Xerox machine tech specs

			DOLPHIN (D0)   		DORADO

IC technology		Schottky TTL		ECL
Address space		4M 16-bit words 	4M 16-bit words 
			(extendable to 16M) 	(extendable to 16M) 
Micro store		4K 36-bit words 	4K 34-bit words 
Microcode cycle time	200 nsec		60 nsec
Datapath		16 bits			16 bits
Main memory		198K - 768K 16-bit wds	512K 16-bit words
						(expandable to 16M)
Memory access time	1 usec			1.4 usec
Cache size		n/a			4K 16-bit wds
						(4-way set associative)
Cache access		n/a			120 nsec
Display			808 x 606 bit map	808 x 606 bit map 
Disk			Shugart SA4008		Trident T-80
			(20M bytes)		(80M bytes)
Net connection		Ethernet		Ethernet 
Cost (internal Xerox)	$15K			$60K

2)  Bill Clancey had GUIDON running on the Dolphin for the demo and
    his subjective feel was like SUMEX under a 3-5 load average.  Thus
    the Dolphin is roughly a KA-10.  He has not run extensively on the
    Dorado to calibrate its feel.

3)  The MIT CADR was demonstrated.  It was hard to calibrate its speed
    since it does not run INTERLISP and hence no benchmark is possible
    for a system we have experience with.  I have the impression it is
    roughly KA-10 speed.  Apparently Richard Greenblatt is still trying
    to set up a company to sell them and he is talking about something
    like half price compared to Symbolics, Inc.  That may or may not be
    real

3)  Peter Szolovits gave me some details on the current Zenith NU
    status.  MIT has 10 of them now.  They use the half speed MC68000
    chip w/o virtual addressing currently.  That will be upgraded as
    Motorola hardware makes it possible.  The machine is designed with
    a 32-bit wide internal bus -- in anticipation of either INTEL's
    coming 32-bit system or the 32-bit Motorola chip.  There are some
    packaging problems in the display and disk that cause overheating
    (these should be easily fixed).  They have an operating system, akin
    to UNIX, just working (it is written in C -- compiler runs on the
    VAX).  Note the difference between this approach and the Xerox
    machine where LISP is the environment w/o underlying operating
    system.  The NU system will support multiple processes.  It will
    come connected to one of the MIT-peculiar networks (not Ethernet).
    It is too early for any benchmarks yet.

Tom R.
-------