Path: utzoo!mnetor!uunet!steinmetz!davidsen
From: davidsen@steinmetz.ge.com (William E. Davidsen Jr)
Newsgroups: comp.sys.ibm.pc
Subject: Re: Extended Memory (and 80286)
Message-ID: <10511@steinmetz.ge.com>
Date: 21 Apr 88 14:47:32 GMT
References: <4710@bcsaic.UUCP> <6853@j.cc.purdue.edu> <1119@maccs.UUCP> <4704@teddy.UUCP> <3890@killer.UUCP>
Reply-To: davidsen@crdos1.UUCP (bill davidsen)
Distribution: na
Organization: General Electric CRD, Schenectady, NY
Lines: 34
Keywords: ps2 model 30

In article <3890@killer.UUCP> chasm@killer.UUCP (Charles Marslett) writes:

| And I thought the '286 was bad til I tried to do some work with the '386.
| Seriously, the '386 is a great chip -- its a somewhat faster '286 with a
| neat feature to speed up MSDOS execution under a real OS.  Mostly, though,
| it is not much faster than an equivalent clock speed '286 in real applications.

  I'm not sure what you mean by "real applications." My experience has
been that most applications from CAD to editors use between 50-70% of
the CPU when recompiled for the 386 instructions. I have seen this for
both FORTRAN and C. For programs using 4 byte ints the speedup is 2:1
most of the time.

  If by "real application" you mean DOS applications using instructions
which will run on an 8 bit 8088, I think you're more or less right.
There are some instructions which run in fewer clocks on the 386, so it
should run a trace faster then a 286 at the same speed, but in general a
286 at the same speed will be about as fast.

  It looks like the 386 works better on slow memory than the 286,
because it does fewer fetches for opcodes, therefore getting hit less by
wait states.  If the 286 takes 2 fetches to get 4 bytes it pays for the
wait states twice (sorry for restating the obvious).  This only becomes
a factor in data when storing data larger than 2 bytes, such as 4 byte
data loads (the P9 should be somewhat slower here, the 286 doesn't do
it) and block moves or string searches. 

  Your figures for branch frequency are very close to mine, and while
there will be places where the longer opcode pipe will be useful, in
general it won't make much diference.
-- 
	bill davidsen		(wedu@ge-crd.arpa)
  {uunet | philabs | seismo}!steinmetz!crdos1!davidsen
"Stupidity, like virtue, is its own reward" -me