> > I hope they're at least running P166s by now.
>
> Well, I'm still running a trio of P100s, as well as a quintet of
> PIIs. They're a damn sight better than nothing; running a LL test on
> an exponent in the 8 million range would be painful, but not half as
> painful as testing a 10 million digit number on a PIII-500!
> (Exclamation, not 500 factorial ;-)
I have quite a few P-166's and they're all doing double-checks (I let
Primenet assign the type of work). I even have a few PPro 200's that are
getting double-checks at the moment since those machines are just busy
enough to make the "rollingaverage" around the 800 or so mark, making it
*seem* like a 160 MHz machine I guess. Oh well, no bother to me.
> > Any word on if Mr. Woltman
> > will be coding a Merced version of Prime95?
>
> Intel will have to release the Merced architecture documentation to
> developers, and George will have to beg, borrow, steal or maybe even
> buy a set of the documentation, and some Merced hardware to practise
> on.
I thought Intel had already released developer info on the Merced, but maybe
they only have the marketing stuff out. Either way, I know that most major
manufacturers have already been given to paper info on the Merced specs,
including electrical info, instruction set, etc. Whether this is public or
not is something else. Like I said, anything I can grab from Intel during
the Compaq conference, I will. If it's not covered by NDA, I'll share what
I can.
> If no-one improves the algorithm, then I'd _expect_ finding a
> teraprime to take about 10^9 times as long as finding a gigaprime.
> There are (obviously) 1000 times as many iterations to do, each
> iteration will take (a bit more than) 1000 times as long to execute,
> and the chance that a single exponent will prove to generate a
> Mersenne prime is only 1/1000 as much.
We're really going to need some factoring code that can do trial factoring
well beyond 2^64, to eliminate as many full LL tests as we can.
What would be a good bit size to trial factor a teradigit prime up to? I
know there's some point in the bell curve to optimize how deep to trial
factor for any given exponent...
> 10^9 is about 2^30, so I'd suggest a timeframe estimate of 30 Moore's
> Law periods between finding the first gigaprime and finding the first
> teraprime. So, something of the order of half a century, assuming
> (and it's a _big_ assumption - the laws of physics are hard to work
> around) that we really can continue to double speed every 18 to 24
> months.
I guess we'll just have to see how relevant the breakthrough's in making
atom width traces and even quantum computers will be to the "real" world.
(to quote: "Reality" is the only word in the language that should always be
used in quotes.)
> I'm 46 now; I expect to live long enough to see a gigaprime (and
> maybe even a 13th human being's footstep on the Moon!), but I very
> much doubt I will see a teraprime in my lifetime, unless there is a
> major, major advance in the theory.
Taking quantum computers to a "real" application might just provide the
paradigm switch you're thinking of, several orders of magnitudes of
improvement *could* be possible in the next, oh...say 25-30 years. Big
"maybe".
Aaron
_________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
