Date: 10 Jul 93  19:35:29
From: Seth Stroh
  To: Tom Mulcahy
Subj: Re: Performance of CISC proces
-------------------
  *In a message dated  7 Jul 93  13:52:51 Tom Mulcahy writes:

TM>  uh why would the fpu in the 040/25 be 3.5 times faster than the fpu
TM> in a 486dx2/66 machine??

  I don't know if it is actually 3.5 times faster but the floating point
processor in the 68040 is significantly faster then that of the 80486.
Here is a few entries from an extensive benchmarks list I got a copy of not
to long ago...(the benchmark was for floating point transcendental
functions)

  Computer                          O/S            Compiler     Result
------------------------------------------------------------------------
  4.7 Mhz  8086 (IBM XT)            MS DOS         tcc 2.0      7300
  7.1 Mhz 68000 (Amiga 500)         AmigaDOS 2.0   Manx cc5.2a  2150
 25.0 Mhz 68030 (Amiga 2000)        AmigaDOS 2.0   SAS C 6.1     369
 25.0 Mhz 80386 (Gateway 2000)      MS DOS         tcc 2.0       318
 25.0 Mhz 80386+80387 (CompuAdd)    MS DOS         tcc 2.0        31
 25.0 Mhz 68030+68882 (Amiga 3000)  AmigaDOS 2.0   SAS C 6.0      10.3
 33.0 Mhz 80486DX (Legatech)        SVR4           UCB cc          6.5
 66.0 Mhz 80486DX2 (Gateway)        MS DOS         tcc 2.0         5.5
 25.0 Mhz 68040 (Amiga 4000)        AmigaDOS 3.0   SAS C 6.2       5.4
 MIPS with FPU accelerator (SGI)    Unix           cc              1.5
 Cray Y-MP C90                      Unicos         cc              0.55

  Just had to stick the SGI and Cray on there!  Anyway, this is a partial
listing from a larger post which (I believe) was originaly off the
internet.  As you can see, in this test which is makes heavy use of
transcendental math (did something like 200,000 sine values and multiplied
them together) the A4000 (a 25Mhz 68040) outpreformed a 66Mhz 80486DX2 but
just barely.  Thats impressive if you consider the difference in clock
rate!  The 486 was more than double the clock rate!  In addition, as I
understand it, the A4000 does not support burst to/from its memory which
may or may not have slowed the 68040 in this test.
  Now if you were to compare a 33Mhz 68040 to the 66Mhz 80486DX2 you would
see an even bigger gap!  The reason for the difference in speed is simply
that the FPU design in the 68040 is simply more efficient then that of the
486.  As I understand, Intel had greatly improved the FPU preformance in
the Pentium.  I wonder if Motorola has made improvements to their FPU for
the 68060???  At any rate, I have similar benchmarks from other sources
that report similar findings...