Author: Dann Corbit
Date: 14:34:02 04/17/01
Go up one level in this thread
On April 17, 2001 at 16:59:20, Bertil Eklund wrote:
>
>First of all these are the CONDITIONS for the match vs Kramnik.
>Use of a SMP-macine with 8cpus. POINT
For which 7 could sit idle. The assumption that 8 CPU's will definitely beat 1
CPU is unproven. Consider ZugZwang, P.ConNers, CilkChess, etc. They were at
Paderborn, running on a titanic pile of CPU's and ... Lost handily to Shredder.
Now, multiple CPU's probably *will* make a properly designed program stronger.
But with 8 CPU's you also have a much greater probability of finding a SMP
coding bug which might get by on a 2 or 4 way system. In other words, the sure
way to find out which is stronger is to test.
>I was asked to recommend the STRONGEST programs for a qualifying tournament.
>The qualifyer should be used to pick the BEST program and not the "luckiest".
>Therefore it should be played in longer matches. Again and I repeat in order to
>minimize the randomness it shouldn't be played in a short tournament in a few
>rounds.
How about qualifying rounds? Start One per continent?
North America,
South America,
Europe,
Asia {if a suitable entrant could be found -- maybe Ren},
Africa {I don't know of any engines here, but there might be some},
Austraila/New Zealand
Then have a face-off between the pairs of winners (with Europe getting a bye in
case of the unlikely event that Africa and Asia enter) -- reason: Europe has
most of the strongest chess engines.
Play the contests over the internet. One match per weekend for a couple months
could create a very large number of games to get a fairly reliable answer.
Maybe we could even use Winboard or the like and play over the internet on ICC
or FICS and have nightly games for two months. That way, many dozens of
entrants could even play effectively.
Now, as we both know, a round-robin tournament is not going to be practical.
BUT, a Swiss tournament does effectively answer exactly two questions:
1. What program is strongest.
2. What program is weakest.
We really have little information about anything between the two extremes, but
those two data points are found with relative reliability (which is the whole
beauty of the Swiss format, and the main reason that it is employed).
>The above means that only a few programs can take part in the challenge.
>What is the strongest programs of today? In my opinion, if I pick the 5-6 best
>programs (one cpu), it can only be the following Deep-Fritz, Deep-Shredder,
>Deep-Junior, Fritz6 and Tiger/Gambit.
That's a pretty good list. I think Ferret might be in that mix also.
>Then I checked the CONDITIONS again, and what is left only the Deeps of Fritz,
>Shredder and Junior. Fritz6 was a bit old anyway, so I skipped it but I really
>was sorry that (after that I had seen that mr Theron wasn't planning any
>SMP-version in the near future) I couldn't recommend the program(s) that may be
>the strongest of all in a single-cpu event.
The conditions of 8 CPU's does not eliminate single CPU programs. You are
making an implicit assumption that 8 CPU programs will always beat 1 CPU
programs. Since these multiple CPU programs are so new, I think that this has
not been demonstrated. In fact, since Shredder trounced all the multiple CPU
programs at paderborn, (if anything) the opposite has been demonstrated. IOW it
is entirely possible for a single CPU program to eat multiple CPU programs for
lunch and spit out the bleaching bones onto the beach.
>
>No there wasn't any ifs included in the stated CONDITIONS.
>No Swiss-tournament in a few rounds against a bunch of GMs.
Instead, you rely on your guess? I don't think you can be sure that it is
accurate.
>IF noone else hasn't noticed it yet, nowadays the most used word in this Forum
>is IF.
Which is exactly as it should be. If we don't know, then we don't. If x is
stronger than B we might be able to demonstrate that IF we play enough games,
etc.
>I have almost seen as many suggestions as there are members in this fora.
>Almost anyone seems to be absolutely sure what to do and how to do.
I think all of the suggestions are better than what was done.
>I only want to remind again that this is a commercial event. Meaning that there
>is a time-schedule. There are rules to be followed and contracts to be signed.
>I also find it quite logical that the organizer of the event want the strongest
>possible program to play against Kramnik.
Or maybe the organizer wants whatever brings in the most dollars -- strongest or
not.
>The cost of the event is a seven
>digit-number in USD.
Wow. I think it remarkable that someone was willing to spring for that.
>At last before everyone starts the shouting and shouting:
>
>I accepted the offer as a consultant, "chess-computer expert" for what it is
>worth, because I really liked the idea of a match between the best program on a
>8cpu-machine.
>
>I haven't received a single cent for this, no the entry-fee wasn't supposed for
>mr Irazoqui or my pockets (mr Czub as usual). I was offered a trip to the event
>but if it can make someone happier I stay at home and prepare the computer for
>the next SSDF hardware-level. Hopefully mr Theron can have some kind of revenge
>if Tiger(s) tops the next list on the new hardware.
For what it's worth, I don't have any feeling that you did anything underhanded.
I do think you did what you believe to be best. Everyone else here would do
things differently, which does not make them right (nor you nor anyone else).
In matters of opinion, there can be a wide range.
How a thing ought to be done is a viewpoint matter. Viewpoints are neither
right now wrong. The only "rub" to the situation is that if the method looks
unfair, you will get a lot of "contrary" viewpoints.
>Bertil
Dann.
P.S.
I am afraid that this is a storm that will never go away. That's on the one
hand. On the other hand, that is probably the best possible thing for the
promoters, since it will mean people will talk about it for years afterwards.
If it were clear-cut, they would forget about it in 6 months.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.