Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!bu.edu!m2c!risky.ecs.umass.edu!umaecs!daly
From: daly@ecs.umass.edu (Bryon Daly, ECE dept, UMass, Amherst)
Newsgroups: comp.windows.ms.programmer
Subject: Re: 486SX
Message-ID: <13597.28283195@ecs.umass.edu>
Date: 8 May 91 17:12:53 GMT
References: <1359452390@p11.f11.n310.z2.fidonet.org> <1024391017@f90.n310.z2.fidonet.org>
Lines: 76

In article <1024391017@f90.n310.z2.fidonet.org>, Alexander_Holy@f90.n310.z2.fidonet.org (Alexander Holy) writes:
>  >  > 486SX == 486 w/o "built in" coprocessor
>  >
>  > 486SX == 386 ?
> 
> No, no. The 486 is a complex 64-bit RISC CPU (very similar to the i860) which 
> is microcoded to do 386 instructions. The hardware is totally different.

The i486 is *NOT* a 64-bit chip (Ref: i486 Hardware Ref. Manual).  It is also 
not a RISC chip.  It is fully a CISC chip, but it does have a RISC core, and
have many RISC-like characteristics (pipelining, etc).  I also wouldn't say the
hardware is totally different.  Instruction Set: same(OK, a few new instr.).
Memory Addressing: same.  Register Set: same.  For most intents and purposes, 
the i486 is a 386 with faster instruction execution times, a cache built-in, 
and a built-in coprocessor.  Take away the coprocessor, and you're basically 
left with a fast 386.
            
> 
> The 486sx seems to be here for marketing reasons, now that AMD has started 

I agree with you here.

> shipping their 386 clone CPUs. I think that the 486sx will perform like a 486 
> of equal speed. The question is, since it lacks the internal FPU, if there will 

Except when doing floating point, when it will be slower than a 386 w/copro.
And except that the 486SX only runs at 20MHz and the slowest 486 'DX' that 
intel makes is 25MHz, for a 20% loss of speed, a leap backwards in technology.

> be a 487sx. If yes, the 487sx/486sx combination is likely to perform 
> drastically worse than a 486.

Here's the joke:  the 487SX is available, but performance of the combo will
not suffer due to interprocessor overhead.  Why?  Because the 487SX *IS* a 
regular 486 (but at 20MHz) and it takes over all functions on the computer from
the 486SX, which can then be removed.  How much for this miracle of marketing
technology?  $200+ MORE than a regular 25MHz i486!  Costs go as such:
~$250 - 486SX (20MHz)
~$550 - 486DX (25MHz), with coprocessor
~$750 - 487SX (20MHz), with coprocessor
so the total 486/7SX combo is almost TWICE the cost of the 486DX, and is
20% slower!  NOT cost effective at all.  If you NEVER buy a copro, you save
$300 big deal.  The system retailer's markup will absorb most of that just 
because it is a "486-class" system. (Not to mention initial costs of designing
new board to accomodate the the combo (they have different pin-outs, of
course)).

> 
> If you are not interested in FPUs, the 486sx may be worth a look. Anyway, there 
> seems to be a need for fast CPUs for doing such weird things as Windows.

Don't do it!  This is just a marketing scam to hurt AMD and their 40MHz 386 
clone.  AMD's competition spells better prices/performance for us consumers, 
and Intel seems to be doing all it can to avoid that.
 
> 
> Clue: If Windows runs much to slow on a 286, and does not make sense to run on 
> a 386sx with less than 2 MB, why are we still using 16bit compilers?

Good question!  Doesn't Zortech C++ do 386 code, though?  I hope MS & Borland
decide to provide a 32 bit option soon.  Perhaps in time for when Windows
goes to the 32 bit flat model, sometime around Windows 7.5? :-) :-)

> 
> Anyone got his hands on the new Watcom 386 C Compiler special Windows Version?

I wish!

Regards,

-Bryon Daly
daly@ecs.umass.edu

----
And no, I don't work for AMD, Intel, or Joe's Pizza.  I'm just a concerned
citizen :) looking for more bang (and less marketing BS!!!) for my buck.