On Thu, 10 Feb 2000, Ramez Naam wrote:
> Hmmm. Having now looked at these projections, I must say that they project
> computing power growth decidedly slower Moore's law.
>
I will simply comment that Moore's law is often misquoted. In its
*original* form (from Moore's original statements in the '60's),
I believe that it *only* discussed circuit *density*. That would
imply nothing about clock speed (advances or lack thereof due to
increases in power consumption), cache or memory bandwidth limitations,
architecture advantages (reduced-instruction-set (RISC),
multi-arithmetic-logic-units/chip, Very-Long-Instruction-Words, etc.).
The net effect of all of these things combined seems to produce
an advancement rate slightly faster than Moore's original projections.
However, in spite of the "Peter's" frequently crying "wolf, wolf",
there seems to be continued progress speeding up the rate of change.
Moravec pointed this out in his recent "Robot: Mere Machine to
Transcendent Mind" book, but the rate of acceleration is so low
that it is difficult to notice. I offered the AMD/IBM observations
as evidence that Moravec's obervations may be accurate.
> Thus, I hope we are not using the fact that we are beating SIA's sub-Moore's
> law projections to argue for near term growth substantially faster than
> Moore's law.
>
The SIA projections are moderately conservative projections (after
all they are an industry consensus). However, there was a "bump"
in their 1993 projections, when things came in a year or so in
advance of the plan around 1996-1997. They passed this off as a
"one-time" occurence in the 1997 Road-Map. It looks like history
is repeating itself with the 1997 map. That would imply that
they may be consistently underestimating the rate of change.
Its going to be interesting to see how they explain things
in the Y2K Road-Map.
Of interest in the links from my pages is the link to the
Petaflops II conference in '99. Some of the papers from that
conference on the advances that are in the prototype stages are
very very cool.
Robert
This archive was generated by hypermail 2b29 : Thu Jul 27 2000 - 14:03:38 MDT