14 Feb
2007
14 Feb
'07
1:59 p.m.
On 14 Feb 2007 at 7:08, arcarlini at iee.org wrote:
It's the fastest clock you can claim to have
anywhere
in your machine while keeping a straight face :-)
...And sometimes, (particularly in the Pentium I look-alike game) it
can mean a number that bears no relationship to any particular
physical quantity.
Consider, for example the Cyrix/IBM Pentium "clones". Rated in "PR"
numbers which was a synthetic quantity that was supposed to resemble
the clock speed that an Intel Pentium *would* have had to operate at
to deliver equivalent performance. Mostly, the PR numbers were sheer
twaddle aimed at a na?ve public.
But then, most benchmarks are nearly useless once vendors get a
chance to figure them out.
In the 80's, a synthetic benchmark for PCs was the Landmark Speed
test number, which attempted to assign a value based on how much
faster one's particular machine was than a stock 5150 PC. You could
go to a show and the silly test would be running on just about every
vendor of Far East clone PCs. I think the Landmark test was mostly
notable for measuring public gullibility.
As an aside, I wonder how many Landmark PC diagnostic kits were
actually sold (I have a couple, complete with alignment discs and
port loopback connectors).
Even well-intentioned attempts at measuring computer performance
across all platforms can run aground.
Sometime in the mid 70's to early 80's, a suite of tests called
Linpack (written in FORTRAN) to measure supercomputer performance.
Once it became known that customers were using the Linpack numbers to
rate processors according to perceived speed, vendors would start
tweaking compilers and then later hardware to deliver higher numbers.
In particular, I remember a startup firm named Saxpy whose machine
architecture was designed to run the Linpack test of the same name
(essentially a sum of two vectors times a scalar constant and
probably less than 50 lines of code). I don't know what became of
the firm, but one of their boxes would certainly be a collector's
item today.
This, of course was nothing new. In the days of big iron, there were
groups within many vendor marketing organizations dedicated to
running potential customer-supplied benchmarks in order to qualify in
the bidding process, usually by any means possible.
Cheers,
Chuck