Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!watmath!clyde!rutgers!lll-crg!lll-lcc!pyramid!ncr-sd!stubbs
From: stubbs@ncr-sd.UUCP (Jan Stubbs)
Newsgroups: net.arch,comp.lang.lisp,comp.lang.prolog
Subject: AI Machine Performance
Message-ID: <1236@ncr-sd.UUCP>
Date: Thu, 6-Nov-86 20:13:27 EST
Article-I.D.: ncr-sd.1236
Posted: Thu Nov  6 20:13:27 1986
Date-Received: Fri, 7-Nov-86 23:28:12 EST
Reply-To: stubbs@ncr-sd.UUCP (0000-Jan Stubbs)
Followup-To: net.arch
Distribution: net
Organization: NCR Corporation, San Diego
Lines: 18
Xref: watmath net.arch:4196 comp.lang.lisp:1 comp.lang.prolog:1

A question for computer architects who dabble in AI. What are the
characteristics of a machine architecture which has good prolog and/or
lisp performance?

Are RISC machines likely to be as good as machines with special hardware
like the Symbolics machines?

Why do the primitive data types I have seen use several bits of each
word as a tag? Wouldn't a separate word (or byte on byte machines) 
give much better performance on currently available hardware at the 
cost of memory space? Is this cost too high?


Jan Stubbs    ....sdcsvax!ncr-sd!stubbs
619 485-3052
NCR Corporation Advanced Development
16550 W. Bernardo Drive MS4010
San Diego, CA. 92127