Message-ID: <1715@inmet.UUCP>
Date: Sat, 18-Aug-84 00:32:41 EDT
Article-I.D.: inmet.1715
Posted: Sat Aug 18 00:32:41 1984
Date-Received: Sun, 19-Aug-84 13:44:15 EDT
Lines: 38

#N:inmet:17900029:000:2040
inmet!muller    Aug 16 21:13:00 1984

From a:muller Thu Aug 16 15:11:12 1984
***
Question (prompted by a Byte article which reviewed Turbo Pascal for the PC...):

The artical was in last month's issue, but it referenced another from Jan,
1983.  In that one, Byte listed benchmark times for the Sieve of Eratosthenes
for many machines, as reported by their readers.  (The Sieve finds the prime
numbers from 0 (1) to 8190, and does this 10 times, I believe.)  In that
Jan, '83 article, there were 2 reportings for Apple Pascal (on a ][ ?), which 
differed by more than a factor of 2; I tried this with the program given in 
last month's article and got results that were slower than the slower of those 
two reportings by a factor of 1.5 or so.  So, what explains these differences?

(1) Perhaps the other readers were guilty of biasing their results.
(2) Perhaps they had Accelerator boards or something, but Byte didn't report
    it or wasn't told.
(3) Perhaps they optimized their programs or their algorithms somehow.
(4) Perhaps there are differences between Pascal 1.0 and 1.1 (I tried this with
    both 1.1 and 1.2 and got identical results).
(5) Perhaps there are differences between the ][, the ][+, and the //e (mine).
(6) Perhaps this reflects loose tolerances in the 6502's clock speed, and
    mine is slow. 

[There was some discussion in Byte about optimizing the program (especially
for different languages) or the algorithm, but this strikes me as invalidating
the whole concept of benchmark comparisons, a messy business anyway.]

Are there any intelligent comments to be made about this?  I am curious about
the timing differences, and the performance of my machine in particular.  I am
not inviting discussion of the general topic of benchmarks, but it just seems 
to me that almost identical machines running the same source code compiled
under the same operating system with the same compiler should execute in about
the same time...unless everything really isn't the same...
    
   Jim Muller       ima!inmet!a:muller (?)  or  harpo!inmet!a:muller (?)