[Prev][Next][Index][Thread]

Re: Executor (was Re: MACINTOSH IS BEST)



NOTE:  I believe this thread has moved to comp.emulators.mac.executor
and comp.os.mac.advocacy, but since my original incorrect statements
were made in comp.os.ms-windows.advocacy and comp.os.msdos.misc as
well as those two groups, I thought my retraction should be posted to
all the original groups.

An earlier discussion comparing SoftWindows for running PC programs on
Macs and Executor for running Mac programs on PCs, contained the
following exchange.

>>>>> "Cliff" == Clifford T Matthews <ctm@ardi.com> writes:
>>>>> "Joe" == Joe Ragosta <jragosta@dca.net> writes:

    Cliff> However, Joe neglected to mention that within Executor's
    Cliff> limitations, an N MHz Pentium will run 68k Mac programs
    Cliff> *faster* than an N MHz PPC601.  In fact, an entry level
    Cliff> Pentium running Executor will run 68k code faster than the
    Cliff> vast majority of 68k based Macs Apple sold, including most
    Cliff> Quadras (ftp://ftp.ardi.com/pub/SynPaper explains how we get
    Cliff> such great performance).

    Joe> ROTFL. I'll tell you what. Try running a PowerMac 7500 with
    Joe> Speed Doubler and System 7.5.3 and I doubt very much that
    Joe> you'll be even in the same ball park.

Joe was right.  Part of what I wrote above was wrong.

The claim that "an entry level Pentium running Executor will run 68k
code faster than the vast majority of 68k based Macs Apple sold..." is
correct.

However, the claim that "an N MHz Pentium will run 68k Mac programs
*faster* than an N MHz PPC601" is no longer correct, assuming the
PPC601 is using Apple's new synthetic CPU or Speed Doubler.  Here are
some relevant benchamarks, with some comments:

                 O   R   I   G   I   N   A   L      N   E   W     estimated
                -------------------------------   --------------
                                                  PPC601  PPC601    scaled
		Quadra	Pentium	486DX4	486DX/2   100MHz  100MHz    Pentium
                  610	 90MHz	 75MHz	 66MHz    w/o SD  w/ SD     100 MHz
		------	------	------	------    ------  ------    -------
CPU		16.018	28.833	15.727	13.840    32.04   54.06     32.036

Dhrystones	19.586	21.886	12.084	 9.424    42.76   54.98     24.317
Tower		18.909	27.130	12.235	11.556    52.00   29.71     30.144
Quicksort	17.759	27.105	15.606	13.919    30.29   57.22     30.116   
Bubble sort	18.409	31.154	19.286	16.875    36.82   57.86     34.614
Queens		19.083	38.167	19.083	18.320    50.89   65.43     42.407
Puzzle		22.083	44.167	23.661	21.032    47.32   73.61     49.074   
Permutations	21.019	28.564	11.604	12.242    48.43   61.89     31.737
Int. Matrix	24.200	26.469	19.369	16.608    70.58   94.11     29.410
Sieve		23.362	60.290	33.982	30.145    31.68   81.26     66.988
		------	------	------	------    -----   -----     ------
Average		20.490	33.881	18.582	16.680    45.641  64.007    37.645


The first four columns are from ftp://ftp.ardi.com/pub/SynPaper, our
white paper that describes how our synthetic CPU is so fast on an
architecture that is so different from the PPC.

The next two columns were contributed by an Executor user who had
access to a 100 MHz PowerMac 7500 on which he ran Speedometer 3.23,
both with and without Speed Doubler ("SD").

The third column is derived by multiplying the second column by 10/9
(i.e. the ratio of 100 MHz to 90 MHz), which we used instead of
getting a 100 MHz Pentium for a few reasons*.

Although the specific claim that "MHz per MHz Executor runs 68k code
faster than a PPC601" is no longer true (it was true when SynPaper was
written, when Apple was still shipping their first 68k emulator), the
point I was making in the exchange between Joe and me is still
correct.  SoftWindows can run a greater percentage of applications
than Executor can, but Executor is significantly faster than
SoftWindows.

I apologize to both Joe and the readers of these groups for my
outdated claim.

--Cliff
ctm@ardi.com

More information, including a downloadable demo copy of Executor, can
be found on http://www.ardi.com/.
__________________
*I wanted to get this apology out ASAP, so that people who catch this
message but didn't see the original exchange will still be able to
retrieve it.

It should be possible to squeak out higher numbers by using a P100
with a faster bus speed and better cache than the Dell Omniplex 590
that we took the original measurements on, but since the PowerMac 7500
isn't in our control, there may be tweaks missing there that would
have resulted in higher numbers there, too.

In addition, the latest "BleedingEdge" release of Executor doesn't get
speedometer numbers this high because of a (soon to be fixed) problem
with our recently added sound subsystem.

Overall it made more sense to just scale the old numbers, admit my
mistake and get back to work on making Executor better!


References: