From: harveyrj@vt.edu (R. J. Harvey)
Date: Fri, 16 Dec 94 09:21:43 PST
To: cypherpunks@toad.com
Subject: something NOT MIME-related
Message-ID: <9412161721.AA08486@toad.com>
MIME-Version: 1.0
Content-Type: text/plain


Hello:
   Not to distract from the entertaining MIME thread, but I've got a 
question that's a little closer to a crypto topic (i.e., software
psueudo-random number generators).  
   In the aftermath of the Pentium-can't-divide-accurately flap, I modified
a random-number generation routine I'd written to check for the presence
of the Pentium divide errors.  In the process, I put in a routine that
did an elementary benchmarking of the chip's performance in both integer
(speed to repeatedly execute an empty for-loop 1 million times) and floating 
point operations (inserting a divide operation in the loop, and adjusting
the resulting execution time by subtracting the time required for the 
empty loop before computing divide-calculations-per-second performance).  
This is an admittedly very crude benchmark, but I wanted to get some 
rough idea how many divides could be performed per minute of program
execution (i.e., to estimate how long the program could run before a
Pentium-problem might occur).
   Anyway, I found what appeared to be very strange results when 
comparing performance on my 486/66 versus a 486/25 and 386/20:  namely,
although the 386 was dead last on both the primarily integer-based empty-
for-loop and for-loop-with-divide timings, the 486/25 and 486/66 turned
in effectively identical times in the empty-loop benchmark (the 486/66 was
about 33% faster than the 486/25 in the divide-based benchmark).  All
machines were running essentially equivalent versions of Windows for 
Workgroups).
   My question is, why would the 486/66 and 25 produce comparable integer-
based empty loop performance?  I haven't tried a comparable program running
under plain-DOS to see if this is somehow Windows related.  I supsect there's
an easy explanation, but it escapes me.  Any suggestions would be greatly 
appreciated.

rj
------------------------------------------------------------------
R. J. Harvey                               (mail: harveyrj@vt.edu)
(PGPkey 0BADDDB5: 82 42 53 EA 97 B0 A2 B2 FC 92 90 BB C2 26 FD 21)