Newsgroups: comp.sys.ibm.pc.hardware
Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!magnus.acs.ohio-state.edu!csn!kessner!david
From: david@kessner.denver.co.us (David Kessner)
Subject: Re: Difference between 386/33 & 486/25 not counting fp
Message-ID: <1991Apr12.093457.4147@kessner.denver.co.us>
Organization: Kessner, Inc.
References: <1991Apr9.085749.4568@agate.berkeley.edu> <1164@gistdev.gist.com> <1991Apr12.073828.20663@agate.berkeley.edu>
Date: Fri, 12 Apr 91 09:34:57 GMT

In article <1991Apr12.073828.20663@agate.berkeley.edu> ilan343@violet.berkeley.edu (Geraldo Veiga) writes:
[stuff deleted]
>Under DOS, QAPlus rates it at 15172 Dhrystones.  Under ISC Unix,
>Drhystone 2.1 (source code from netlib, compiled with gcc)  gives a
>reading of ~~ 17000.  This is the best I can get, with all memory
>addresses cached.
>
>Can different motherboard designs give such radically improved
>performance?  I suppose the more likely explanation is that the
>benchmark code is not the same.  The only change I made to the
>Dhrystone code was the HZ constant used in the timing routine 
>(from 60 to 100).
>
>Can anyone enlighten me on this one?

It would seem that the the dhrystone code really isnt the same.  My best guess
is that the 'other guy' used Dhrystone 1.x on the 486 systems-- which does
rate 486/33's at 34-35,000 Dhrystones.

I also suspect that you were using a rather lousy compiler in generating
you dhrystone results...  But that is just a guess.

Doing any benchmark while running under DOS (even with a DOS-extender) really
cripples the test-- since the program must be lobotomized to run in that 
environment.  I appluad your try at UNIX...

Anyway, when doing tests like this it is important to use similar/same 
compilers/options on the two CPU's-- a luxery we seldom have.  That's why we
loot at the various trade journals.

Personal Workstation, Unix World, and UNIX Review has rated all 486/33's in the
ballpark of 34-37,000 dhrystones.  All 486/25's in the 23-27,000 Range.  The
386/33's get 15-18,000.  And the 386/25's get about 11-13,000.

Since these benchmarks are all ran in very similar environments (possibly even
the same binary), it's fair to say that the results are reliable.  When noting
that there were no stray figures (like a 486/33 doing 17,000), then it's even
easier to have faith in the figures...

Anyway.  These figures would indicate that the 486 is twice as fast as the 386
for the same clock speed.

It is also interesting to note that:

	The 486 can be sped up further by re-arranging the instructions so
	that they make better use of the 486's parallelism/pipelining.  None
	of the UNIX compilers make use of this "well documented by Intel"
	feature.  Actual milage may vary, but all indications are that the
	486/33 could do 40K+ dhrystones with this optimization.

	The dhrystone does not test floating point performance.  All
	indications show that the 486 is about three times as fast as the
	387-- about 1.5 MFLOPS.  This is pittifully slow when compared to
	other CPU's like the 040 (3.5 MFLOPS), but I can live with it...

Well... That's my two cents worth...  

-- 
David Kessner - david@kessner.denver.co.us            | do {
1135 Fairfax, Denver CO  80220  (303) 377-1801 (p.m.) |    . . .
If you cant flame MS-DOS, who can you flame?          |    } while( jones);
