Path: utzoo!utgpu!news-server.csri.toronto.edu!torsqnt!lethe!yunexus!rreiner
From: rreiner@yunexus.YorkU.CA (Richard Reiner)
Newsgroups: comp.sys.ibm.pc.hardware
Subject: Re: C64 emulation.
Message-ID: <22590@yunexus.YorkU.CA>
Date: 28 Apr 91 14:50:27 GMT
Article-I.D.: yunexus.22590
References: <1991Apr21.041312.1@watt.ccs.tuns.ca> <22498@yunexus.YorkU.CA> <1991Apr22.035236.25021@agate.berkeley.edu>
Organization: York U. Computing Services
Lines: 38

c60b-1eq@e260-1e.berkeley.edu (Noam Mendelson) writes:

>In article <22498@yunexus.YorkU.CA> rreiner@yunexus.YorkU.CA (Richard Reiner) writes:
>>a cached 386-33 does a best-case 13500 dhrystones/sec (according to my
>>own measurements: dhrystone 2.0 compiled with MS C 5.1, generating
>>286-specific instructions, using the register option).  Thus a 386-33
>>is 375 times as fast as a C64.                                 ^^^^^^
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
>I know this is a bit off the track but that statement is simply wrong.
>Comparing a 386-33 to a C64 using dhrystones is ridiculous, considering
>the difference between the processor capabilities.

The Dhrystone benchmark is *intended* for the comparison of different
processors, and even processors of different architectures.  The
introduction of the documentation packaged with the Dhrystone 2.1
source reads, in part,

	    it gives a first performance indication which is more
    meaningful than MIPS numbers which, in their literal meaning (million
    instructions per second), cannot be used across different instruction
    sets (e.g. RISC vs. CISC).

Now, the comparison of an i386 to a 6502 in integer and logic
performance alone (i.e. in what Dhrystone measures) may be ridiculous,
since this omits the more sophisticated capabilities of the i386
(memory management, etc), but my statement about the relative
performance of the two processors is not wrong: it does *not*
constitute a misuse of the benchmark, and *does* give a perfectly
reasonable estimate of how much faster an i386 can be expected to
perform logic and integer ops.

There are so many benchmarks around that extreme care must be
exercised in interpreting them correctly.  Careless statements, like
the one above claiming that my comparison is "wrong," only contribute
to the confusion.

//richard