Re: [sl4] Call for information: Moore's Law

From: Tim Freeman (tim@fungible.com)
Date: Mon Jul 14 2008 - 12:41:13 MDT


From: "Thomas McCabe" <pphysics141@gmail.com>
>...Currently, we're looking for information on what causes Moore's
>Law-style exponentials in computer hardware, so we can predict what
>will happen given conditions different from those of the past forty
>years....Any links to other theories, or literature on the subject,
>would be greatly appreciated.

I always assumed that a random innovation takes a random amount of
time and causes an improvement by a random small percentage, and
there's a probability distribution on those percentages and times. If:

* many small innovations happen, and
* there aren't any physical limits that skew your probabilities,

then the law of large numbers takes over, so you have predictable
exponential progress that depends only on the probability
distributions.

I don't have a link for that. It always seemed obvious.

The above hypothesis does predict that spending more money so you get
more small innovations would give you a larger growth rate for the
exponential growth. I haven't looked at the numbers. If the research
spending has been fairly constant, and we got a constant exponential
growth rate, then that's consistent with my theory. If the research
spending is growing exponentially, and we get a constant exponential
growth rate instead of something super-exponential, then that's
inconsistent with my theory.

>1)., Moore's Law was a self-fulfilling prophecy because it was used
>to set industry goals,

That's a conspiracy theory. Each competitor would benefit by beating the
prediction, and none of them do, for no particular reason. Doesn't work.

>2)., most technological growth is exponential, because the rate at
>which technology improves is proportional to the amount of technology
>already in existence.

That seems to me to predict super-exponential growth, since I'd expect
a random research result to give you a random percentage improvement,
and improved technology should help you get research results faster
from one year to the next.

Maybe the research is inherently sequential? If it is only possible
to discover small improvement nubmer n+1 after completely understanding
small improvement number n, and there are humans in the loop with a
limited learning rate, then you could expect Moore's law to hold no
matter how much money you throw at the problem.

-- 
Tim Freeman               http://www.fungible.com           tim@fungible.com


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:03 MDT