Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka-Zappa Information Center
- - By Dadi Jonsson (Silver) [is] Date 2007-09-22 16:15 Edited 2007-09-27 19:24
Rybka vs Zappa. Mexico City, 20-27 September 2007

** This post will be kept up-to-date during the match with all relevant information (corrections and additions are welcome!)

Introduction

Clash of the Computer Titans. Press release from Convekta.
Rybka arrives, police is up in arms.
Rybka-Zappa press conference (video). Very interesting. Source: ChessVibes
Europe-Echecs: Match Rybka-Zappa (video)

Live broadcast

Official site.
Web broadcast by Convekta. Shows evaluation of current position by Rybka. Nice design.

Games are also broadcast on the ChessPlanet server. A new version of the client has been released which adds a window in which Rybka constantly monitors the game in infinite analysis mode (also used for the WCC-2007 games). More information here:
On-line Rybka evaluations and here.

Final score

Rybka: 4,5
Zappa: 5,5

Games

The match will consist of 10 games, with a time control 60 minutes per game plus 20 seconds per move.

Games start at 15:30 UTC/GMT. (10:30 local time in Mexico, 11:30 AM EDT (US east coast), 17:30 CEST).

When two games are played on the same day, there will be a 10-15 minute break between the games.

20th Sep – Game 1: Pre-game info (Dagh) -- Zappa - Rybka 1/2-1/2
21st Sep – Game 2: Pre-game info (Dagh) -- Rybka - Zappa 1-0
22nd Sep – Game 3: Zappa - Rybka 1-0
22nd Sep – Game 4: Rybka - Zappa 0-1
23rd Sep – Game 5: Zappa - Rybka 1-0
24th Sep – Game 6: Rybka - Zappa 1/2-1/2
25th Sep – Game 7: Zappa - Rybka 1/2-1/2
26th Sep – Game 8: Rybka - Zappa 1-0
26th Sep – Game 9: Zappa - Rybka 1/2-1/2
27th Sep – Game 10: Rybka - Zappa 1/2-1/2

Zappa wins the match with 5,5 - 4,5.
Parent - By Jim Walker (***) [us] Date 2007-09-23 13:02
The 10:30 EST may confuse some people since U.S. East coast is still on daylight savings time.  So 10:30 Mexico time is 11:30 on the East coast of the U.S now.
Parent - - By drd (*) [us] Date 2007-09-27 13:36
So where are the last 2 games?    This post is not being kept up to date as claimed
and I would like to see the 9th round game.     I cannot find it anywhere on the web.
Parent - By Felix Kling (Gold) [de] Date 2007-09-27 13:48 Edited 2007-09-27 14:01
Sorry, I'll update it.

Btw., you also can find all the games here: http://www.chessok.com/broadcast/live.html
Parent - By Gryphus (**) Date 2007-09-27 14:48
that the force is with Rybka today
:)
Parent - By Uly (Gold) [mx] Date 2007-09-27 19:15
It ended 4.5 - 5.5 as foretold by Intagrand!: 

> Given many believe the odds are strongly stacked against Zappa, I'd bet that Zappa wins with 5.5-4.5.


> I've seen a couple of examples where Zappa Mexico's evals were more accurate than Rybka's and Zappa was
> able to win. I've conducted no long tests, but those examples were enough to convince me that Zappa can
> defeat Rybka if the game strays into those kind of positions.


> Given the quality of Rybka's book, it will be unlikely, but I would still rather put my money on Zappa with
> better returns if the unlikely happens

Parent - By joshua2 (*) [us] Date 2007-09-28 04:33
I want more info on the computers. Were both computers equal the whole time? What was the clock speed of the 8-core? How many kns? How does this compare to a quad core overclocked? I think we can do comparable by giving say 33% more time for an overclocked quad and it will be same "speed"?
Parent - - By Ricardo_dk (*) [dk] Date 2007-09-28 06:41
Hi, i dont uderstand why does Rybka chess program lost the match
So I think i will buy zappa program and wont rybka????
Thanks
Parent - - By revengeska (**) [us] Date 2007-09-28 08:10
You must realize that the Mexico match was held on enormously fast hardware, and this is why Zappa made up the ground vs Rybka.  If you're running a normal computer at home, Rybka will still most likely be stronger, and the various rating lists available reflect that.
Parent - - By Ricardo_dk (*) [dk] Date 2007-09-28 08:11
Thanks for your message.
Do you know when Rybka 3 is coming out?
Bye
Parent - By revengeska (**) [us] Date 2007-09-29 02:36
I believe it's set to come out at the end of this year.
Parent - - By isilverman (***) Date 2007-09-28 11:39
Why did Rybka agree to an 8 processor match, then?
Parent - - By josh [us] Date 2007-09-28 11:58
No disrespect to Vas and the rest of the Rybka team but , I for one am glad that Zappa won the match. Maybe now Vas is hungry once again.
That wasn't just Rybka playing at Mexico but the pride of the strongest.
Parent - By josh [us] Date 2007-09-28 12:00
No doubt the book was a huge factor but a blemish for the Rybka team.
Parent - - By FICGS (**) [fr] Date 2007-09-28 12:01
Definitely it would be interesting to know how many nodes per second were calculated by both programs on this 8 cores hardware, and what Rybka team expect about future improvements of Rybka on this issue.
Parent - - By ernest (****) [fr] Date 2007-09-28 12:14
Thibault,
Il a déjà été expliqué x fois que le mode de calcul des nodes par Rybka est "différent" de celui des autres programmes (dans le code, il y a quelquepart une division par 14, ou une bidouille de ce genre...). Les nodes ou les nps ne sont donc pas comparables.

En plus du fait que de toute façon, on ne peut rien conclure de la force d'un programme par son nps.
Parent - - By Svilponis (***) [ee] Date 2007-09-28 23:43
It would still be interesting to see what was maximum kn/s on this 8 core Mexico monster. I'm mostly interested of Rybkas nodes. It is not perfect base for comparing engines performance, but still better than nothing. I think kn/s gives still pretty good overview about hardware performance. If my modest 2 core system makes with Rybka usually between 200 and 300 kn/s, then I'd expect something near 1000 kn/s of the hardware used in Mexico.
Parent - By Linus (***) [at] Date 2007-09-29 07:16

> then I'd expect something near 1000 kn/s of the hardware used in Mexico.


On Playchess my modest C2D (250-300 kN/s) has been playing against other Rybka 2.3.2a's which displayed up to 600 kn/s. I assume that most of those were highly o/c'd Quads. Your estimation of near 1000 is probably right.
Parent - By FICGS (**) [fr] Date 2007-09-30 10:57
Je ne voulais pas parler du nombre de nodes / sec. , mais de la comparaison du nombre de nodes / sec. de chaque moteur entre 1 proc et mp... Pour voir la "perte" en rentabilité... Car au final, s'il s'avère que cela revenait à faire courir Zappa sur une machine 2 fois plus rapide, le résultat parait logique...

--

I did not mean the number of nodes / sec. that is not interesting indeed , but the comparison of this number under single processor & mp. for each engine... If this match was like running Zappa on a twice faster hardware, the result is quite logical...
Parent - By revengeska (**) [us] Date 2007-09-29 02:37
I don't think the Rybka team really thought Zappa had real chances.  It's well known that Rybka doesn't scale well at 8 cores, Vas himself can tell you this.
Parent - - By Quapsel (****) [de] Date 2007-10-02 11:11

> Why did Rybka agree to an 8 processor match, then?


Why should have Rybka not agreed with an 8 processor match?
Because they have fear? The chances are too bad?
Because they want to tell the world "we are strong, but only on a small machine!"
?

The guy who wants to be accepted as world-champion shouldn't shout
- the boxing gloves have to be green
- the boxingring has to be red
- and the gumshield of my opponent has to be yellow.
He should win and 'good it is!'.

No, IMO that decision was right, Rybka ist strong enough to do a match on every hardware.
And she has had a good chance, maybe her chance was larger than that of Zappa.
But Zappa has won. Engine, Book, luck? Who knows at this Moment.
Maybe private larger matches on such hardware will show more clear.

Rating-Lists createt with smaller machines are a lot in the net.
A result made on a hardware of the near future is the much more interesting thing, I think.

Perhaps Vas will have a look upon the parallel-Features of Rybka, maybe, he will do something.
Perhaps he thinks "No, other things are more effective.
Vas will know best then!

Quap
Parent - - By Hetman (*****) Date 2007-10-02 18:17
Hi,
I agree that the programm has to have to be universal one to some extent.
For me the rating calculated of the mostly used machines is important because the rating on 64 processors
monster I will never (5 year period) buy has no meaning for me.
Regards
Hetman
Parent - - By Quapsel (****) [de] Date 2007-10-04 07:11
ACK. But:
A few years ago I stupidly thought "Wow, machines with two Processors! Who would buy such a strange thing!"
And today even rather cheap PCs don't do it with only one core.

How many years, and a normal kiddy-gaming-Machine will have 16 Cores?
next year 4 Cores, in three years 8 and in 5 years 16?
Maybe, the PC you will buy in 5 Years will have such a core-Farm inside.
OK, 64 Cores might come 4 more years later, our over-next Machine :-). (OKOK, only speculation)

What think experts about it? When will come 16-Core-CPUs for normal Users who want CPU-Power for any reason?

Quap
Parent - By Hetman (*****) Date 2007-10-04 18:33
Yes, the computer technics evolution is very quick. Because of that I have given restriction in 5 years.  ;-)
I think that in 5 years period I will not need stronger PC but after five years I will be not able to buy single processor machine.
I want not to be a victim of the race for the processing power. I would like to have the pleasure of playing the chess and current level of Rybka is ok, from that point of view, the small adjustments in style.
The current view of topchess is not nice for me: game  15-20 moves from the book, then 5-6 moves from home computer analysis and then own play is starting. I think that future is in FRC and we all will be grateful to RJ Fischer.

Rgds
Hetman
Parent - By ExperiMental (**) [us] Date 2007-09-29 14:17
I was surprised to see Rybka lose to Zappa, and now I'm thinking maybe this is a good thing.  I think Vas has done amazing work with Rybka, and I suspect this will motivate him to win a rematch.  If Vas does go for a rematch then he will be going all out, in code, and we will end up with the results.
I love the Rybka program that I purchased.  I don't even have the latest version, but the version 1.2f that I use is still awesome.  I have run so many computer versus computer matches and this version of Rybka beats all other engines that I own.  In my testing there has been no contest.  For me, Rybka rules!
Parent - - By masomusic (**) [nl] Date 2007-10-01 17:42
Very unexpected outcome, as Zap(pa) on most testmatches seems at least 100 elo lower than Rybka. I didn't fancy the change of engine version(s) during the match of Rybka. It's obvious Rybka 2.3.2a could have done the job well by itself. From ratinglist even at 4 cores Rybka would have better chances, as the engine is superior from opening till far middle game to any engine on market today. The endgame needs improvement that's obvious and everybody knows that who used Rybka in past. Vas often needs to hear things twice to take something serious I know :-) (just for old time sake, Vas, I still luv you) It didn't feel right to me that Zappa won the game and it's for sure a bullshit outcome. Well, maybe it's for the better, while all attention goes to Zappa now, Rybka has time reveal herself again. Or something. I had a personel giggle when I noticed Zappa Mexico being sold at Shredder's website, I guess dear Stephan had nothing more to offer this year, hehe
Parent - - By turbojuice1122 (Gold) [us] Date 2007-10-01 18:25
It's obvious Rybka 2.3.2a could have done the job well by itself.

Rybka 2.3.2a would have gotten creamed.  The result would have been on the order of 7.5-2.5, I think.
Parent - - By masomusic (**) [nl] Date 2007-10-01 19:17
Only if there would have been some sort of an 8 core bug in the 2.3.2a engine I guess...was there? Also I can't imagine Zap(pa) benefitting so much from more cores if it's "chessknowledge" is not on par with Rybka's. On 1 core machine Zap(pa) gets creamed as you can see from the ratinglists. Can someone explain how it's possible Zap(pa) gains so much more strength compared to Rybka once the core amount doubles?
Parent - - By turbojuice1122 (Gold) [us] Date 2007-10-01 22:55
Not only were the engines used, except in game 1, superior to Rybka 2.3.2a in terms of multiprocessor scaling, they were noticeably better at playing a game of chess (at least, on average--for an exception, see game 9 with 71...f4??).  Furthermore, Zappa Mexico was largely tuned, especially in terms of the opening book and the types of positions it plays well out of the opening book, against Rybka 2.3.2a; using that version in Mexico would have been absolute suicide.  I would have bet my life savings that using that version against the Zappa version that Erdo used in Mexico would have lost the match by a larger amount than 5.5-4.5.  Zappa Mexico on 1 CPU gets creamed on the rating lists because not only are the conditions typically shorter time controls, which no Zappa version enjoys, but because the rating lists are biased toward engines that play well in random situations, while a specifically tuned engine (even if it's not done that way on purpose) like Zappa gets screwed.  That's why we have matches like in Mexico to even things out.
Parent - By masomusic (**) [nl] Date 2007-10-02 19:15
Nice explanation turbojuice1122, you might be very much right about it. If indeed the Zappa team prepaired so well and studied Rybka 2.3.2a's weaknesses for a long time it for sure had some good chances to to some severe damage. The thing is, how much time one needs to really outsmart an engine by preparation. Hard to guess I think. Anyways, I think you re not far from truth.
Parent - By Hetman (*****) Date 2007-10-04 18:19
Elo it is not all. I think to much attention is given to that numbers. Elo is based on Gauss statistic, might be the Poisson statistics would be better or the other one :-). Chess are not Brown movements.
Rgds
Hetman
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka-Zappa Information Center

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill