[Beowulf] commercial clusters

Robert G. Brown rgb at phy.duke.edu
Fri Sep 29 13:34:45 PDT 2006


On Fri, 29 Sep 2006, Vincent Diepeveen wrote:

> Actually with respect to chess, it is not overblown,
> there is 1 billion people on this planet that can play chess and the average 
> IQ of a chessplayer
> is significantly better than the world wide IQ (you can ask yourself whether 
> this is caused by
> chess, or whether it is the people who choose to play chess that are clever).

You have to be very careful to differentiate between "can" play chess,
as in know or at some point in their life knew the rules of chess, "do"
play chess as in have played at least one game of chess in (say) the
last three months and "are obsessed" with playing chess (as in their
last game was less than 24 hours ago no matter when you ask them.

> Nearly everyone who has an university degree or is high in government, can 
> play chess.
>
> Logically that this includes many persons with a lot of money.

At this point, I'm guessing that a desktop computer with an
over-the-counter or open source chess program (configured to play its
best game) can beat 99% of that billion handily and consistently.  Maybe
even 99.9%.  Remember that it includes many, many people who "can" play
chess but in fact do play chess only rarely, or last played ten years
ago, or who play but are just not very good.

Of the perhaps million or so who remain, many of them very likely enjoy
chess as a social thing as much as for the game, or find equal-class
human opponents via the internet when they wish to play and are at their
computer.  When you get into this million or three people, they probably
CAN find better players online than most computers are currently capable
of, and playing against those players is basically free (fixed cost of
internet connection in the first place plus maybe some trivial fee).
Most of them DO probably play down in the park in the evenings, or at a
club, or with their neighbor down the hall, or with a husband, wife,
child, good friend.

I personally think that the number of people who would pay anything much
in order to play a computer (cluster) online is pretty minimal, and if
they ARE very good, the cluster is going to have to work pretty hard to
offer just one person a decent game at a time from what I read (although
you'd know more than I would about that:-).  Not sure that this is a
viable commercial market on a lease basis, but differences of opinion
are what make entrepreneurial risk exciting, right?

Go is where the real challenge is, though.  Computers {\em can} play at
the level of a chess grand master.  From what I've read, though, a small
child with only a few years of experience can beat any Go-playing
computer ever built, often beat it badly.  Go is so complex that it has
been asserted that there have never been two identical games of Go
played by humans and likely never will be (consider the permutations of
a 19x19 board filled in in alternating binary with complex rules
governing captures and the acquisition of territory.  There are ballpark
10^{10^50} possible games in Go compared to a mere 10^50 or thereabouts
possible positions in chess.

The amazing thing is that humans can PLAY Go and play it well.  It is
massively nonlinear and nonlocal, but humans seem to be able to "see"
strategies that will succeed at the tactical level AND balance the "give
some to get more" progress of a successful game out to advantage.

If anybody could figure out how to abstract the critical information
theoretic content of successful go and reduce the problem from where it
is now, with a search space that simply cannot meaningfully be explored,
to a game not based on tree searches at all but rather on some higher
order vision of game complexity, and build a good Go playing computer,
now that might sell.... and if it didn't, you'd be famous anyway.  Fame
is worth something:-)

> But you have to *offer* them something that is attractive to buy. A good 
> deal.
>
> Not some electric heater that's gonna sink 3 kilometer into the ocean (and if 
> you checkout
> the new SSE4 instructions it's obvious that intel will let the itanic sink), 
> soon after you buy it.

Naaaa, you just have to wait for the first large scale quantum
computers, which will do coherent parallel processing of correlated
phase information qubits to solve in a matter of seconds at least SOME
problems that take much, much longer on bit crunchers.  Since the
computation proceeds coherently, the evolution process is reversible (no
entropy loss) and so no heat is generated from when you load the problem
to where you "measure" out the result.  If you're really lucky, the
qubits are itty bitty quantum dotty things as well and you can pack a
lot of them into a very small volume;-)

     rgb

>
> Vincent
>
> ----- Original Message ----- From: "Jim Lux" <James.P.Lux at jpl.nasa.gov>
> To: "Geoff Jacobs" <gdjacobs at gmail.com>; "Vincent Diepeveen" <diep at xs4all.nl>
> Cc: <beowulf at beowulf.org>; "Angel Dimitrov" <stormlaboratory at yahoo.com>; 
> "Robert G. Brown" <rgb at phy.duke.edu>
> Sent: Friday, September 29, 2006 5:18 PM
> Subject: Re: [Beowulf] commercial clusters
>
>
>> At 06:56 AM 9/29/2006, Geoff Jacobs wrote:
>>> Vincent Diepeveen wrote:
>>> > If it was possible to build your own cluster in easy manner and then > 
>>> run
>>> > for
>>> > example a chessprogram at it in a user friendly way,
>>> > there would be 100k+ clusters right now of 64 cpu's and more.
>>> 
>>> I think you maybe overestimate the number of chess players who can
>>> challenge existing chess programs on their standard PC. I also think you
>>> overestimate the number of chess players with ~$10k to blow on a cluster.
>> 
>> Vincent's example of chess players may be a bit overblown, but his point is 
>> very well taken.  There are an enormous number of people who could use "big 
>> computation" if it were "easy to use" and "cheap enough".  $10K is a 
>> significant price point.  Aside from traditional consumers of such things 
>> (research labs, etc.) $10K is low enough that it is almost a consumer item, 
>> albeit probably not mass market in the millions of sales context.
>> 
>> 
>> 
>> Jim
>> 
>> 
>

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu





More information about the Beowulf mailing list