Path: utzoo!utgpu!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!bloom-beacon!think!ames!pacbell!belltec!jim
From: jim@belltec.UUCP (Mr. Jim's Own Logon)
Newsgroups: comp.sys.ibm.pc
Subject: Re: 386SX for PC/AT
Summary: 386SX and whether it is faster.
Keywords: 386SX, AT
Message-ID: <310@belltec.UUCP>
Date: 20 Dec 88 15:21:51 GMT
References: <1031@logico.UUCP> <1550@micomvax.UUCP> <12835@steinmetz.ge.com>
Organization: Bell Technologies, Fremont, CA
Lines: 34

>   I agree with that, however, if you buy a 386 compiler such as Phar
> Lap, you should see a large improvement in performance. On an 80386
> changing from 286 to 386 instructions gives 2-3 improvement in most
> cases, I would expect about 2:1 faster for a 386SX.
> -- 
> 	bill davidsen		(wedu@ge-crd.arpa)
>   {uunet | philabs}!steinmetz!crdos1!davidsen
> "Stupidity, like virtue, is its own reward" -me

    The simplest 386SX upgrade is a daughter card that plugs into the 
286 socket. Runs at the same clock rate. Still is 16 bits. Still has the
same performance (within a few percent). Doesn't matter which compiler you
use, the hardware fetch of instructions is the same rate. And 16 bit code
is 16 bit code (this does sound a lot like a hardware geek putting down
the software, but in this case the point is true). A 386 compiler will not
buy you any more than a few percent improvement because of the instructions
that a 386 has and a 286 doesn't have.

   The original poster had the right idea for real speed up. The SX runs 
at the 16MHz speed of the CPU and has a local cache to support that speed.
Only for cache misses (you will be lucky to get anywhere near 90% for UNIX)
do you slow down for the regular speed of the 286 memory fetches. A company
already makes a board just like this. Aox Inc. of Waltham Mass., no phone
number in the article. My favorite line in the add/article: "Cost of the
board is determined case by case". 

   Anyway all of the speed stuff is right from the Intel application notes.
Running at the same speed of the 286, the most you'll see is 15% speed up,
and when you talk to real Intel engineers they say that is a max theoretical
number and the real number is more like 5% for real applications.


							-Jim Wall
							 Bell Technologies, Inc.