Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!rutgers!ames!ucbcad!ucbvax!decvax!decwrl!pyramid!oliveb!sun!pajaro!gil
From: gil%pajaro@Sun.COM (Gilman Chesley)
Newsgroups: comp.arch,comp.lsi
Subject: uP Figure of Merit
Message-ID: <15252@sun.uucp>
Date: Wed, 18-Mar-87 14:06:50 EST
Article-I.D.: sun.15252
Posted: Wed Mar 18 14:06:50 1987
Date-Received: Fri, 20-Mar-87 02:39:28 EST
Sender: news@sun.uucp
Lines: 15
Keywords: microprocessors, performance, figure of merit
Xref: mnetor comp.arch:616 comp.lsi:66

I don't want to flog a dead discussion (Kwh per Mhz) but with the advent of
the Am29000 the need still exists to evaluate single chip microprocessors
independently of the process generation, so I would like to suggest a figure
of merit for microprocessors:   DHRYSTONES*microns^2.

The basic argument is that to normalize for process generation one should 
adjust for density, which would then give a rough figure of merit for
architecture/implementation. For example, Figures 2 and 8 of Myers, et al,
MICROPROCESSOR TECHNOLOGY TRENDS, Proc. IEEE, Dec. 1986 show a good
correlation between the increase per year in the number of microprocessor
transistors and MIPS: a constant 10 IPS per transistor. Presumably an
architecture significantly over that ratio is more effective than one
significantly under that.

Gil Chesley {usual disclaimer}