Path: utzoo!utgpu!water!watmath!clyde!att!pacbell!ames!amdcad!sun!pitstop!sundc!seismo!uunet!portal!cup.portal.com!mslater
From: mslater@cup.portal.com
Newsgroups: comp.sys.ibm.pc
Subject: Re: Intel 386SX chip & its applications
Message-ID: <6859@cup.portal.com>
Date: 25 Jun 88 17:17:28 GMT
References: <206900116@prism> <6734@cup.portal.com>
Organization: The Portal System (TM)
Lines: 36
XPortal-User-Id: 1.1001.4222

> | You can build (and I'm sure several companies will sell) a piggy-back board
> | with a 386SX and three PALs, which will plug into a 286 socket.  However, this
> | won't buy you any speed -- it just gets you 386 software compatibility, and
> | in fact will be a bit slower than a 286 for many applications.

> That's misleading.  Since the AT bus can be run at original speed
> while the P9 runs at 16 MHz, I would assume that all piggyback boards
> will run at that speed.

> More to the point, I question that the 386 will be slower in any case. 
> The clocks for any instruction seem to be comparable, and the 386 has a
> longer pipeline, so it should run slightly faster in linear code with
> slow memory.  Note I don't claim this for any other conditions. 

A simple piggyback board will use the clock from the 286 socket, and will
thus run not at 16 MHz, but at whatever speed the 286 was running.  Thus, I
think my original statement is correct.

As for 286 vs 386 speed, I've seen a number of benchmarks that show that for
16-bit code (8086 or 286 code) the 286 runs very nearly the same speed, and
sometimes slightly faster, than a 386.  There are many instructions that
require fewer clocks on a 286 than on a 386.  The primary advantage of the
386 is that it can fetch two 16-bit words at a time; however, this apparently
is not such an advantage, presumably because the pipeline can be kept nearly
full with only 16-bit fetches and because at every jump there is a 50/50
chance that the second word fetched must be thrown away.  The 32-bit bus of
the 386 obviously has a big advantage for programs that manipulate lots of
32-bit data words, but most programs are dealing with bytes or 16-bit words.

If anyone else has any insights on 286 vs 386 performance, I'd appreciate any
clarifications.  The data I've seen indicates that nearly all the performance
gain of 386 systems over 286 systems, when running DOS or OS/2, is simply due
to the faster clock, and that the 32-bit bus is of little help.

Michael Slater,  Microprocessor Report    415/494-2677
mslater@cup.portal.com       sun!portal!cup.portal.com!mslater