Greengard's ChessNinja.com

Rybka Wins 2008 Comp WCh

| Permalink | 31 comments

The program Rybka has been the strongest in the world for years now, according both to the comp-comp rating lists and the opinion of just about any GM you ask. The latest confirmation of this came, sort of, at the computer world championship in Beijing, where chess was part of the Computer Games Championship. (Apparently Beijing is hosting the world championship of everything this year. This was surely the first comp tournament held at a golf club.)

I say sort of because, as usual, the programs playing in the event were running on wildly disparate hardware. According to this ChessBase report, Rybka was on an incredible 40-core machine. It's hard to talk about a program's superiority when it could be on a machine that is twice as fast as those of most of its competitors. But there are good arguments for playing these comp-comp events with the best hardware the operators can get their hands on. Some are more valid than others. The fundamental one is that not all programs can run on the same hardware. Others, like Hydra, are inseparable from their custom hardware. So you'll never have exact, or even approximate, platform equivalency anyway. Then you have the "best chess possible" argument, which says these events are about assembling the strongest possible chess creature and not about settling a practical argument about which program is stronger on a commercial platform.

Which of course is what interests most casual users and even the afritzionados who live and die with their thousands of comp vs comp games. Most people want to know what program is strongest on their home PC, which isn't going to be 40 cores any time soon. Any program is strong enough to easily brutalize any human not in the top 100 these days though, so quibbling about which program is 3050 and which is 3100 is a little comical. This makes inroads into style and positional understanding by programs the real focus today, which is what makes Rybka much more than just a high elo pretty face. It has a relatively slow search and is not the fastest tactical engine, but it more than compensates with superior knowledge. This has the potential to make it a much more effective teaching tool for amateurs, which was exactly the stated goal of Rybka's programmer, IM Vasik Rajlich.

Speaking of Hydra, what ever happened to it? Google puts a "this site may harm your computer" warning on the links to its official site in the UAE, oddly enough. I suppose its team sort of ran out of challenges, although it has lost to human-machine combo players in advanced chess events.

ChessBase, the Rybka guys, and other comp chess luminaries like David Levy have been casting around for some way to sustain interest in human-machine chess, which has been a meal ticket for 20 years. Now that the Rubicon has been crossed into computer dominance, the goal is to find a way to make these matches competitively interesting without distorting the game entirely and/or giving humiliating odds advantages to the humans. Ideas? Or is comp-machine a thing of the past?


Mig, several experts (C. Donninger - Hydra, A. Cozzie - Zappa) have expressed an opinion that Rybka in fact is incredibly fast searcher. Cozzie said, after seeing Strelka's source code (which is widely considered to be reverse-engineered Rybka) told that it is the most optimized top program.

The number of nodes it displays in the output window is just the real number of nodes divided by 10 or 8 or something like that.

They tried hard to create a myth about THE smart chess program.

I would pay $20 to see Anand or Kramnik (whoever wins the coming match) go 12 rounds with Rybka, standard human tournament rules with no handicap, just to see how exactly the machine smashes the human #1.

This means setting a high contempt factor to grind to (sometimes) bare kings, leveraging the PC's advantage in stamina and tactical non-oversight.

Would also pay $20 to see it play in Corus A (2009 edition), in like manner.

HCL Right now I dont see the Rybka match being remotely attractive to a top GM unless wads of cash available not withstanding your $20 start :-)
The top guys have too much trouble with each other to devote serious attention to a massive prep effort to defeat a Silicon monster. Until another Kasparov emerges this is likely to be the case.

How about 6 simultaneous matches between the top 10 humans and Rybka with $1 million to the highest points scorer. The problem here maybe that Kramnik would be strongly favoured to play very solidly for draws and rack up the half points. At least with this format the top GM's may not lose out to their rivals by devoting huge energy to the prep.

I'd be very interested in the preparation the likes of Gazza and Vlad would have put in as compared with a WC match. I think it was v Deep Junior that Gazza donned funny goggles with cash inducements (couldnt see that happening in any of his various KvK encounters)

Yah, a tournament would be more interesting. Few remember that Junior played in Dortmund in 2000 and got an even score despite losing two horrible games in old-school anti-computer style. A fairly strong IM/GM closed event in Argentina included a computer for a number of years but it won by increasingly wide margins. But it would be interesting to see top GM modern anti-computer play. (Preparation + profit motive.) I'm convinced that a strong GM can play for a draw with good success at classical time controls. It's just understandably hard for them to swallow their pride and stop playing for a win...

Good result of Hiarcs was yet amazing, I would say Rybka might have bee a bit lucky to win, because draw with Hiarcs was quite likely result at this level. Like said, Hiarcs is quite old program but still peforms incredibly well. One might argue that if Hiarcs would have the same processing power available than Rybka did, it might have squeesed that draw, but then again it's only speculation...

The programming style of Hiarcs is also very interesting and it is even possible that when the computing power increases Hiarcs might be the one that benefits most out of it, because deeper the search goes more heavy it is to calculate by brute force and thus more intelligent program gets more and more advantage, since the difference between the depth gets smaller and thus likelyhood of Hiarcs making a tactical mistake in visible horizon diminishes.

Hey Mig I think the problem with a tournament is that the top GM's wouldn't put in the necessary preparation. If Rybka was added to Corus A a SuperGM would be better off spending his preptime on games against other SuperGM's rather than perhaps having to radically alter style etc to prep against a program.

To me the story of the man vs. machine challenge is over. It's too one-sided now.
The next story should be automated evaluation and explanation. It's fine to have a chess program that beats me 100 times out of 100, but I would like that can explain to me what I do wrong in my human-to-human games. Right now, the only thing it can explain to me, are tactical blunders.
There is a lot of potential, but I'm not sure it can be done with the current mindset of chess programmers. They focus on speed and node counts, but not so much on user interfaces and how to deal with human-understandable concepts.
Maybe programs could play tournaments in a special category where they have to lay out plans, which are then carried out by identical modules. You would have to invent a formal representation for chess plans. The identical playing modules would try to stick to the plans as much as possible - just like humans do, without the blunders.

Oops, didn't see the top comment. You guys are fast!

Either Rybka's search is so fast that it has transcended to a point at which speed simulates knowledge or it is actually smarter. I say this not from reverse engineering code but from looking at its play and analysis. From the beginning of its existence it was clear it had a very different, and usually superior, evaluation function. By this I mean it was closer to agreeing with GM evals of material imbalances and pawn structures. Show several engines positions that aren't responsive to search depth to test their evals and they often disagree dramatically.

As supporting evidence, I would include the fact that Rybka has always been a relatively inferior rapid tactician. E.g. scoring worse on standard tactical test suites than engines it dominates in head-to-head play. This isn't very relevant on a practical level, but if it scores worse on test suites and then beats the same engines, it suggests something besides search optimization is going on. But the quality of its eval is the main thing, and it's tough to say how this would be achieved via pure speed unless it was several generations beyond the rest.

"Any program is strong enough to easily brutalize any human not in the top 100 these days though, so quibbling about which program is 3050 and which is 3100 is a little comical"-Mig
"Any super GM is strong enough to easily brutalize any human not in the top 50 these days though, so quibbling about which super GM is very great and which is even greater is a little comical"-Mig edited by chesshire cat. Myabe this argument could preempt the Moro discussions etc :) ?

Either Rybka's search is so fast that it has transcended to a point at which speed simulates knowledge or it is actually smarter.

What if it has all the knowledge an average GM has (like Rybka's programmer), plus algorithms gained from data mining the entire chessbase database, plus a farther search horizon, plus whatever ideas gleaned from GM consultants coded as part of its engine?

Cuz that's what we may be looking at. Rybka's GM author has coded his chess knowledge into the program. It's no longer some wimpy 1600-elo player or near-novice coding blind (think Fritz, Chesstiger, and Junior's authors).

Dude, what are you talking about. First of all Rybka's programmer is an IM not a GM. Second, of course all major chess programs have access to high-level 'chess knowledge', from GMs, databases, books, whatever. The problem is how do you translate this 'knowledge' into program logic (the eval). Clearly, this is where Rybka has found a better, perhaps fundamentally different way.

Oops, I mistakenly thought Rajlich made GM recently. I just double-checked at the fide site. (Btw, his wife outrates him by 100 pts, he-he-he.)


Anyway, how does a 1600-elo programmer code chess knowledge he doesn't really understand? The world was long-waiting for a Rajilich with the programming skills and chess ability in the same body...

This suggestion probably comes too late, as the comps are already too strong to really make it interesting... but anyway:

I would like to see a man vs machine match in which the program would make its own preparation. No human-made opening book, no playing around with contempt values between games. Give it access to chessbase, tell it who the opponent will be and let it work out the rest by itself. Perhaps it could somehow be sealed off from human influence for a set amount of time before the match.

This would of course require some extra programming (probably a lot), but the interesting thing is that it puts into play some of the fields in which humans are relatively strong: Psychological preparation, capacity for adapting to unexpected circumstances and so on.

i would much rather see fischer random chess with no opening books for the comp, than any sort of material or times odds match for the human.

since no human being can be expected to develop an opening book for the 960 starting positions, why allow a computer to. my guess is the top players would crush rybka.

bondegnasker, that's what happened more or less in the last game of the Kramnik-Deep Fritz match: Kramnik deviated from theory very early and Fritz started to play in a strange way, with quite a non-human plan which, however, got kramnik on the ropes very early. See the comments by yusupov in some magazine(NiC, I think). Kasparov also wondered if the programmers weren't actually hindering their own computer's play by "forcing" them into a human opening book. All I can say is that i find that game (and the comments) fascinating.

What we need now is for Rybka (or some other machine), to give a simultaneous exhibition against 500 amateurs, or 10 GMs. One machine, one program, simultaneously playing many other players.

Now that would be interesting.

The rules of chess do not allow for referencing books during the game. Just turn off all opening and endgame books to give a huge advantage back to the humans.

My guess is humans would use their opening book knowledge to gain an advantage that the computers would challenge with their tatical abilities. It would also create more closed games and strategy would the human's trump card with the computers trying to equalize with tatics.

Serouisly how much is computing power and how much is encylopidia brain-bashing. Anyone want to memorize the LA phone book and then compete with google for search speed?

Turn off all books and just make it computing power. NO BOOKS PERIOD.

It is against the rules of chess to touch a piece and not move it. I know that my rating against computers is increased when I use the takeback move. Let the players have one take-back move or see the computers next move to a suggested move. Either suggestion would increase players strength of play against computers.

The Rybka node counts are low by a factor of least 15, maybe 20 in some positions. The displayed count is essentially the number of internal White make-moves divided by 7. Null moves are also not counted, which puts the factor above 14. Depth is differenced by 3 from the internal. As the Zappa guy said, "Who cares? It still plays good chess." The "high knowledge" is somewhat of a myth, particularly if by it you mean superior evaluation, at least until IM Kaufman came on board, as even Rybka 2.3.2a didn't use much beyond the "Fruit-like" evaluation of mostly mobility.

If Hiarcs (and perhaps Shredder and Junior) would finally enter the "modern world" of 64-bits, they'd switch to bitboards. This alone would not make them competitive on the rating lists, however.

In reality, I find all this "knowledge" talk is a bit strange, as Rajlich himself says that he simply programmed what gave the best results. The Rybka team's testing methods seem to the best around, currently.

What happened to Hydra? You must have missed this article, published a week ago:

Fast search doesn't mean the computer sees tactics well. In fact, speed has little to do with it. What matters is what part of the search tree is pruned during search. If you want to be tactical, you have to prune less. Rybka searches very fast, and prunes a lot.

There have been simple experiments playing Fruit vs Strelka on depth=1, that is-with no search on both sides. Only eval. Fruit wins convincingly.

Anthony Cozzie gave a very good way to figure out whether the engine relies on search or on good eval. If the evaluation often makes jumps from one depth to another, you have an engine with heavy positional eval. If the eval instead changes incrementally, then you have a fast searcher.

Rybka, at least the 1.0(beta) version, has a lousy eval, very good search heuristics and is lightning fast.

There is something good lost when these chess engines compete on non-identical computer hardware.

Playing on massive 40 core servers has the obvious benefit of resulting in better moves.
But there would be something nice about seeing the engines compete on common dual core computers that any one of us could reasonably have in our home.

Can Kramnik beat the plain Fritz_11 I have on my single CPU computer? Often special non-released versions of these chess programs are used. And specialty hardware is frequently used. It all leaves me unsure.

I believe each August in Mainz (ChessTigers.de), the chess960 computer competitions use modest and identical hardware.

GeneM, you can checkout any number of computer chess rating lists. those usually use identical hardware and the commercial version of the engines. Rybka is still numero uno in those.

It seems that Rybka is Fruit with some improvements (a clone just as are Toga and Sterlka for instance) and V. Rajlich will end up charged with plagiarism.
The case is still 'sub judice' but evidence is mounting ominously ( as of sept. 2008)

see more at :


More on Rybka-Fruit controversy

Parts of Rybka code appear to have been copied verbatim from Fruit (the chances for such thing happening unintentionally are essentially zero).

I don't agree at all, Rybka is also computer world chess champion in Fisher Random 960, so your statement that human can beat it in this variant sounds like something next to impossible.

"i would much rather see fischer random chess with no opening books for the comp, than any sort of material or times odds match for the human.

since no human being can be expected to develop an opening book for the 960 starting positions, why allow a computer to. my guess is the top players would crush rybka."

Sorry, no chance. Rybka would crush any human - probably even more convincingly than in standard chess with opening books. The big amount of theory helps leveling the field in standard chess - if the human doesn't even have that, no reason to think they won't be even more humiliated. That is, they will probably be outplayed starting from move 1 instead of move 15!

Some time ago, I did need to buy a good car for my business but I did not earn enough cash and could not buy anything. Thank goodness my mother suggested to try to take the loan from reliable creditors. Thence, I did that and was satisfied with my collateral loan.

It would be more than interesting. Thanks and keep up your work. It's a good post.

I visited this page first time to get info on people search and found it Very Good Job of acknowledgment and a marvelous source of info......... Thanks Admin! http://www.reverse-phone-look-up.net

Plagiarized stuff gets you nervous? Don't get angry! If you heard about plagiarism checker, then you would better use it! In fact, I do not like plagiarized contents as well!

Twitter Updates

    Follow me on Twitter



    About this Entry

    This page contains a single entry by Mig published on October 9, 2008 5:27 AM.

    Awesome Augury Action was the previous entry in this blog.

    WCh 08 Team Spirit is the next entry in this blog.

    Find recent content on the main index or look in the archives to find all content.