Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!seismo!rutgers!ames!sdcsvax!ucsdhub!hp-sdd!hplabs!sdcrdcf!trwrb!scgvaxd!wlbr!etn-rad!jru
From: jru@etn-rad.UUCP (John Unekis)
Newsgroups: comp.sys.ibm.pc
Subject: Re: Intel Microprocessors (fair benchmarking)
Message-ID: <264@etn-rad.UUCP>
Date: Fri, 28-Aug-87 17:30:44 EDT
Article-I.D.: etn-rad.264
Posted: Fri Aug 28 17:30:44 1987
Date-Received: Sun, 30-Aug-87 09:00:21 EDT
References: <4381@intelca.UUCP> <416@aucs.UUCP> <4048@utai.UUCP>
Reply-To: jru@etn-rad.UUCP (0000-John Unekis)
Organization: Eaton Inc. IMSD, Westlake Village, CA
Lines: 29

>
>In conclusion have you noticed that the micro wars ended abruptly when
>the 386 bencmarks were published.  I guess this is because people who
>have something to proove have no more fiction to go on.  What we have
>in the preceeding message is a futile attempt to throw some sand
>in the form of 68030 into our eyes.
...
  I see that you did not notice my counter -posting to the bench marks.
  The tests in byte magazine produced extremely misleading results, to the
  point of being completely falsified. What was timed by the authors in byte
  magazine was the delta time between the start and stop messages their
  program printed out. Since they were running under a special 386 protected
  mode software package, the start message was probably significantly delayed.
  This gave the impression that the software ran in seconds, when the actual
  elapsed time was probably over a minute. To confirm this I ran their 
  fibonacci sequence routine on a 16 Mhz 80386/80287 and on a 12 Mhz 68020
  with 68881. The 80386 was running UNIX V in 32 bit mode, and so was the
  68020. The timing was done with a UNIX time utility which measures actual
  CPU usage, as versus the BYTE article which used a hand held stop-watch.

  Under these much more accurate conditions the fibonacci test ran
  in 69.1 sec on the 12Mhz 68020/68881 and in 60.3 sec on the 16Mhz 
  80386/80287.   Even with a 25% clock speed advantage the 80386 was
  only 12% faster.(BYTE said the 80386 would do it in 3.1 sec) 

  The conclusions are that accurate benchmarks are hard to perform,
  that you should always double check results on other machines,
  and that any test which shows that much of a performance lead 
  for INTEL has obviously been RIGGED.