Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!usc!snorkelwacker!mit-eddie!bu.edu!bu-cs!mirror!prism!rob
From: rob@prism.TMC.COM
Newsgroups: comp.sys.ibm.pc
Subject: Re: Speed in MIPS
Message-ID: <206900162@prism>
Date: 19 Jan 90 15:26:00 GMT
References: <111607@<1990Jan10>
Lines: 38
Nf-ID: #R:<1990Jan10:-11160700:prism:206900162:000:1681
Nf-From: prism.TMC.COM!rob    Jan 19 10:26:00 1990


gillies@p.cs.uiuc.edu writes:

>386 (25)		5.4
>486 (25)		14   = 2.6*5.4

>I think there is an extreme element of '486 marketing bullshit/hype in
>this last performance figure.

   Can't fault your cynicism, but the 14 MIP figure for the 486-25 is
fairly well founded as these things go. It's based on several measures.

   1) Intel claims the average instruction execution time for the 486 is
      1.7 cycles (vs. 4.4 for the 386). This gives 25/1.7 or 14.7 MIPS. 

   2) The classic '1 MIP' machine is the VAX 11/780, as measured by the
      dhrystone. The 11/780 scores about 1500 on the dhrystone. The 486-25, 
      according to figures published in Mips magazine, scores from about 
      16,300 to 26,000, depending on the OS/compiler combination used. This
      gives a 'VAX-MIP' range of about 11 to 17.

   3) The standard MIP-rater in the MS-DOS world is the Power Meter
      MIPS rating. It tends to be a bit on the high side (it rates
      cached 386-25's and 386-33's at about 6 and 8 MIPS, respectively).
      I've seen at least one claim of over 15 MIPS for a 486-25 from 
      that benchmark. 

   4) Most CPU bound benchmarks I've seen show the 486 to be about 2
      times as fast, clock-per-clock, as the 386. Given a 5-6 MIP figure 
      for the 386-25, that puts the 486-25 at about 10-12 MIPS. This
      is running 8086 code with no 486 (or 386) optimizations.


   Given all this, a figure of 14 seems defensible, though an argument could 
be made for figures lower or higher than that.    

   On a related note, Motorola is now claiming 20 MIPS for its long-awaited,
but as yet unavailable, 68040 (not clear what clock speed that's measured at).