Xref: utzoo comp.sys.ibm.pc.hardware:5377 comp.unix.sysv386:4779
Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!natinst!dell!chs!sauer
From: sauer@chs.dell.com (Charlie Sauer)
Newsgroups: comp.sys.ibm.pc.hardware,comp.unix.sysv386
Subject: Re: SPECmarks....
Summary: Intel probably wasn't first to separate fixed/floating SPEC results
Keywords: benchmark specmark speed ratings
Message-ID: <1991Feb11.153610.20207@dell.dell.com>
Date: 11 Feb 91 15:36:10 GMT
References: <1991Feb8.201336.24388@ux1.cso.uiuc.edu> <10230@ncar.ucar.edu> <1991Feb11.073342.381@ico.isc.com>
Sender: @dell.dell.com
Distribution: comp
Organization: Dell Computer Corporation
Lines: 24

In article <1991Feb11.073342.381@ico.isc.com> rcd@ico.isc.com (Dick Dunn) writes:
>> ...Others complain however, that is more weighted in favor of
>> floating-pt. performance vs. integer performance, so Intel branched off
>> and created the I Spec or Integer Specmark.
>
>"Others" is approximately "intel" - to the extent that pundits say that "I
>SPECmark" really means "Intel SPECmark".
>
>The serious political point in this is that SPEC hasn't blessed anything
>called an "Integer SPECmark" (as far as I know).  The serious technical
>point is that, if the SPEC test suite overemphasizes floating-point (which
>I personally think it does, but what do I know?), the "I Spec" ignores it
>completely, which is rather worse.

Maybe I didn't see the first Intel documents which separated the 4 nominally
integer benchmarks from the 6 nominally floating point components of SPEC 1,
but the first public use of this distinction that I remember came from John
Mashey of MIPS, shortly after the RS/6000 announcement.  Of course, John 
properly presented the individual results, the geometric mean of the floating
components and the overall geometric mean, as well.
--
Charlie Sauer       Dell Computer Corp.        !'s:uunet!dell!sauer
(512) 343-3310      9505 Arboretum Blvd        @'s:sauer@dell.com
                    Austin, TX 78759-7299