Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!think.com!zaphod.mps.ohio-state.edu!mips!apple!equinox!jimi!reed!whitney
From: whitney@reed.cs.unlv.edu (Lee Whitney)
Newsgroups: comp.sys.ibm.pc.hardware
Subject: Re: 486SX - Intel now telling lies
Message-ID: <1991May29.153843.27750@unlv.edu>
Date: 29 May 91 15:38:43 GMT
References: <1991May14.054200.28095@ux1.cso.uiuc.edu> <21020.282f6b40@oregon.uoregon.edu> <1991May14.200856.1431@ux1.cso.uiuc.edu> <1991May15.210339.17118@unlv.edu> <1561@aoa.UUCP>
Sender: news@unlv.edu (News User)
Organization: UNLV Computer Science and Electrical Engineering
Lines: 38



When I posted the first article concerning Intel's claims on 486sx-20
performance, I was only skeptical, but I just read the first
486sx/486dx/386dx performance comparison's from an unbiased source (PC
Magazine), and it does not do much to defend Intel's credibility.  I
really want to believe Intel, so if anybody can explain why these PC
Mag benchmarks are so low, please save me from distrusting ANOTHER
manufacturer.

Just to recap, Intel claims the 486sx CPU is 40% faster than a
386DX-33.


From PC Mag June 11:  '..even though our tests use 32 BIT INSTRUCTIONS,
the faster operating speeds of the 386-33 and the 386-40 propel these
chips PAST the 486sx.'

It turns out that the AM 386-40 was actually 32% FASTER than the best
486-sx machine they tested when running a 386 instruction mix, and the
386-33 was 7% faster.  Keep in mind this is a CPU only benchmark and
does not average in disk I/O, graphics performance, etc.

Here is the Intel justification for this: (PC Mag June 11) 'It's
because DOS's 8-bit addressing limitations cause overhead as
instructions are sent in 8-bit segments and must be converted into 32
bits and then back to 8-bit segments.'

Yeah right.  Even though the above statement does not make sense,
let's just pretend it does.  Even if it was true, can it explain their
47 percent error (claimed + 40%, actual - 7%) ?  Not to mention the
excuse for the slowness was address translation overhead, which they
claim the 486 is SUPPOSED TO BE BETTER AT because of the more tightly
integrated MMU! 


I have no logical explaination for this, and if anyone else does, I
will rescend my accusation of Intel lying.