Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka Cluster Rental Program
1 2 3 4 5 6 7 Previous Next  
- - By Vasik Rajlich (Silver) [pl] Date 2010-12-30 18:56 Edited 2011-02-22 19:02
Visit the website of the Cluster: http://cluster.rybkachess.com/

Overview

The Rybka Cluster is a high-performance supercomputer located in Minden/Westfalen, Germany. Combining software and hardware engineering, the Rybka Cluster features exclusive Rybka versions not available to the public running on customized hardware built with the specific goal of providing top Rybka performance at an affordable price.

Starting on February 1, 2011, the Rybka Cluster will be available to the public for the first time under the Rybka Cluster Rental Program.

History

The Rybka Cluster was initiated in September 2008, when an initial version was built just in time to participate in the 2008 ICGA World Computer Chess Championship in Beijing. The first results were promising and steady improvement in both software and hardware ensued. By December 2010, the Rybka Cluster had evolved into a polished chess analysis tool.

Cumulative figures (as of December 2010) include:

- 2,000 hours of remote client logins
- 4,000 hours of direct cluster testing
- 60,000 hours of software-simulated cluster testing
- 12 major computer chess tournament wins in 12 attempts, including 2 ACCA World Championships and 3 ICGA World Championships

Timeline:



Wins 16th ICGA World Computer Chess Championship in October 2008.
Wins 28th Dutch Open Computer Chess Championship in November 2008.
Wins 11th CCT Tournament in March 2009.
Wins 14th ICGA Open Hardware Computer Chess Olympiad in May 2009.
Wins 17th ICGA World Computer Chess Championship on 8 physical cores in
Obtains positive score in 10 out of 11 informal mini-matches against top human centaurs in the summer of 2009.
Wins 3rd ACCA World Computer Rapid Chess Championship in August 2009.
Wins 29th Dutch Open Computer Chess Championship in October 2009.
Wins 4th ACCA Americas' Computer Chess Championship in November 2009.
Finishes with a positive score in all 11 informal mini-matches against top human centaurs by November 2009.
Wins Mundial Chess Freestyle tournament in February 2010.
Wins 10th International CSVN Tournament in May 2010.
Wins 4th ACCA World Computer Rapid Chess Championship in July 2010.
Wins 18th ICGA World Computer Chess Championship in September 2010.
Wins 30th Dutch Open Computer Chess Championship in November 2010.

A collection of all public games played by the Rybka Cluster can be found here.

Hardware

A computer cluster is a collection of individual computers which act together as a single bigger computer. The challenge is to split the work in such a way that the relatively slow communication between the individual computers is tolerable. When this is accomplished, clustering provides a flexible and relatively inexpensive way to create and maintain a high-performance supercomputer.

The Rybka Cluster, designed, owned and overseen by Lukas Cimiotti, is built from a large collection of electronic and infrastructural equipment, including twenty-nine motherboards, six hundred fifty-six gigabytes of RAM, five terabytes of solid-state disks, fifty-eight CPU sockets, and two hundred ninety-six physical Intel Nehalem cores. A gallery can be found here.

Due to the flexibility afforded by the clustering concept, the exact specifications of the Rybka Cluster change on a regular basis as our team makes improvements.

Rental Protocol

Under the Rybka Cluster Rental Program, anyone can purchase time on the Rybka Cluster in continuous chunks of at least 48 hours. Customers receive client software as well as personalized account information and then use this software and account information to connect to the cluster from anywhere and at any time during their paid-for rental period. When a customer is logged in, the cluster presents itself on his client computer as a standard (UCI) chess engine, which can be installed and used in any standard chess GUI.

Features

The Rybka Cluster includes all of the standard features expected of a chess engine, such as multi-variation analysis, exclude-moves analysis, pondering, full state reset, game play supporting all possible time controls, and so on. All 4-man and 5-man tablebases as well as selected 6-man tablebases are installed.

The Rybka Cluster is also robust to the various connectivity problems which clients can experience. When a client loses his internet connection, the last analysis he requested continues to run, and when he reconnects the accumulated analysis is automatically made available to him.

Finally, the Rybka Cluster has a highly developed and fully configurable notion of draw avoidance. This is useful in computer chess tournaments, to make sure that the cluster's ultra-deep searches don't result in overly cautious moves leading to too many draws. Our clients have also found the feature useful for preparing opening variations which may be objectively equal but which yield good practical winning chances.

Policies

There are no restrictions on what a customer can do with his or her Rybka Cluster time. Customers are welcome to use the Rybka Cluster for their own private chess analysis or automated game play, or to share their Rybka Cluster time with others, either informally or for profit.

Rates

Two standard, well-tested Rybka Cluster configurations are available starting February 1, 2011:

A Rybka Cluster 40 consists of 40 physical cores and can be rented for 238 Euro (incl. VAT) per day, or 200 Euro (without VAT) per day for customers outside the EU or businesses outside Germany. The minimum continuous Rybka Cluster 40 rental time is 5 days.

A Rybka Cluster 100 consists of 100 physical cores and can be rented for 595 Euro (incl. VAT) per day, or 500 Euro (without VAT) per day for customers outside the EU or businesses outside Germany. The minimum continuous Rybka Cluster 100 rental time is 2 days.

Custom configurations are also possible. Our current capacity is 296 physical cores, and this figure can be increased if necessary. Please don't hesitate to contact us about alternative possibilities if you are interested.

Confidentiality

All aspects of Rybka Cluster rentals, including any discussions with us about the matter, will be treated as confidential by our team. Only Lukas Cimiotti and Vasik Rajlich will be privy to any such information.

Contact Us

Lukas Cimiotti is responsible for both the Rybka Cluster and the Rybka Cluster Rental Program. Time reservations and payment arrangements should be made directly with him. The preferred method for contacting Lukas is via private message at the Rybka Forum, addressed to the nick "Lukas Cimiotti". Impressum is here.

Long-Term Reservations

Rybka Cluster time is rented on a first-come first-served basis. We accept reservations for up to six months into the future, and recommend that you make important reservations safely in advance to guarantee availability. Whenever we upgrade our hardware, advance reservations will automatically be upgraded to configurations of equivalent value.

Alternative Rental Possibilities

The main limitation of the Rybka Cluster Rental Program is that cluster time must be reserved in advance and in continuous chunks of at least 48 hours. ChessBase GmbH and Convekta Ltd. are now working on solutions which will allow their customers to access the Rybka Cluster at any time without advance booking on a per-minute basis. We'll provide information about these possibilities when they become available.

Additionally, cheaper Rybka Cluster configurations of less than 40 physical cores will be made available later in 2011.

Disclaimer

Time on the Rybka Cluster is offered as is. The Rybka Cluster has been extensively tested in a wide range of scenarios, but we make no warranties of any kind, either express or implied, including but not limited to the implied warranties of performance, merchantability, satisfaction, or fitness for a particular purpose.
Parent - - By Nelson Hernandez (Gold) [us] Date 2010-12-30 19:16 Edited 2010-12-30 19:28
Congratulations on finally getting this launched.  I did the math, and if you wanted to run the 100 package from Germany at 595 euros per day, and extrapolated that rate to the entire 296-core cluster for a full year, it comes at the current FX rate to about $850,000.  What a deal!
Parent - - By Fulcrum2000 (****) [nl] Date 2010-12-30 19:22
Yep, and now full speed ahead for the the product, the Rybka PCI version. I hope... :smile:
Parent - - By Vasik Rajlich (Silver) [pl] Date 2010-12-30 19:42
Actually I did my schedule yesterday for 2011 and this won't make it, even with my optimistic scheduling. :smile: We're looking at 2012 now for that.

Vas
Parent - - By David Evans (***) [gb] Date 2010-12-30 19:55
You got to be kidding my Vas u have either gone insane or lost all grip on the state of the worlds finances.......

If anyone can afford the above good luck to them.
Parent - - By Felix Kling (Gold) [de] Date 2010-12-30 20:00
This is just the first step before having a more flexible rental model. However, the above is quite some hightech offer, it's like renting deep blue or something like that. No one can expect it to be cheap :)
Parent - - By tomgdrums (****) Date 2010-12-30 22:14

> This is just the first step before having a more flexible rental model. However, the above is quite some hightech offer, it's like renting deep blue or something like that. No one can expect it to be cheap :)


Nor can we expect timely bugfixes to the products we did purchase.  :)
Parent - - By mdraith (**) [de] Date 2010-12-30 22:36
The more time the rybka team gets, the more bugs will be eliminated. Do you have a date and want to impress your spousy with the new rybka release? :lol:
Parent - - By tomgdrums (****) Date 2010-12-31 00:16

> The more time the rybka team gets, the more bugs will be eliminated. Do you have a date and want to impress your spousy with the new rybka release? :lol:


Nah.  I just am becoming more and more amazed at the business model here.

Hiarcs (and consequently Junior) are much better about getting fixes and updates out to their customers.  As is Shredder.

The Rybka 4 book (and R3) is a one off deal with no updates.  The Hiarcs book gets updated 3-4 times a  year.  The Junior book got updated with no extra cost.

Rybka has no way to use it's book internally in GUIs other than Aquarium or Chessbase..  Harcs and Junior (junior to a lesser extent) do have a way to use their book internally.

And now after everyone has spent their money on R4, instead of a timely (I think 7 months after release is not asking that much) update we get an announcement of a rental plan that very few can afford.

All of that adds up to fairly bad customer service.

Perfection is not what I require.  What I require is some sense that the business cares about the customer...that is all.

Instead what I see is that the Rybka team is moving on to different type of customers.  And that is their prerogative.  I see where the money I have spent on Rybka has gone and it ain't gone far.
Parent - - By Felix Kling (Gold) [de] Date 2010-12-31 00:20
common, you can't compare those opening books with Jiri's book. It's no pure engine opening book, of course it's not the best choice for engine tournaments anymore, but you get a lot of opening theory in one compact book, which is quite good for tournament chess. With your argument any written opening book is bad, since it's always outdated :)

Rybka can use the Aquarium book with the book adapter in any GUI, so it's like an internal book.

The cluster thing was planned to be released more than 14 months ago (you may remember the announcement on chessok.com). It will be started soon and then the next thing will be the update.
Parent - - By tomgdrums (****) Date 2010-12-31 00:32

> common, you can't compare those opening books with Jiri's book. It's no pure engine opening book, of course it's not the best choice for engine tournaments anymore, but you get a lot of opening theory in one compact book, which is quite good for tournament chess. With your argument any written opening book is bad, since it's always outdated :)
>
> Rybka can use the Aquarium book with the book adapter in any GUI, so it's like an internal book.
>
> The cluster thing was planned to be released more than 14 months ago (you may remember the announcement on chessok.com). It will be started soon and then the next thing will be the update.


I can of course compare those books.  The Hiarcs book is great and is also meant for studying.  And it is constantly updated.

And your point that somehow my statement makes any written opening book bad is incorrect.  Written opening books (the good ones anyway) also have verbal explanations of the ideas etc. etc.  So even if a move or two is "outdated", the written books still hold value because of their explanation of the ideas.  I am only comparing engine books to engine book regarding value.  So stay on track please.  And the Hiarcs book is great and very deep and wide!  I use it for study all the time.

And I have tried the adapter thing and could never get it to work.   Why does their need to be extra steps?  Why doesn't Rybka just make an internal book?   Or would that be too much like good customer service??
Parent - - By Felix Kling (Gold) [de] Date 2010-12-31 00:39
Internal books have certain disadvantages, for example the visual thing :)

Why didn't you get the adapter working? just copy it to your aquarium folder, open the configuration GUI, choose engine and book ... and voila
Parent - - By tomgdrums (****) Date 2010-12-31 00:42

> Internal books have certain disadvantages, for example the visual thing :)
>
> Why didn't you get the adapter working? just copy it to your aquarium folder, open the configuration GUI, choose engine and book ... and voila


I agree with the visual thing.  Which is why Hiarcs' solution of providing both is great.

I tried the adapter three times and just never got it to work.   Probably my fault but I say again why does it need to be so many extra steps?  Just make it available as an internal book as well?
Parent - - By Felix Kling (Gold) [de] Date 2010-12-31 00:43
those steps aren't many. It's three steps: 1) extract the adapter 2) change the settings 3) install it as engine

It has the advantage that you can edit such an "internal" book.
Parent - - By oudheusa (*****) [nl] Date 2010-12-31 11:40
i have also never been able to get the book adapter to work in the chessbase GUI's.

are you certain it is worth one last shot?
Parent - By Lukas Cimiotti (Bronze) [de] Date 2010-12-31 13:12

>are you certain it is worth one last shot?


It certainly is. For me it works without problems. I have it in the Aquarium folder. Engine should point to the engine you want to use. For me this looks like this:
engine=C:\Cluster\Rybka 4v3 master.exe

And there must be a book section like this:
[Book 0]
config=Cluster3
maxmoves=99
Parent - By tomgdrums (****) Date 2010-12-31 14:30

> i have also never been able to get the book adapter to work in the chessbase GUI's.
>
> are you certain it is worth one last shot?


I for one am not going to give it one last shot.  The only engine I would need it for would be Rybka.  Every other engine I use (including the free Stockfish) allows me to use an internal book for playing the Shredder GUI.  So once again the Rybka model supplies a lot less usage for the money.
Parent - - By Felix Kling (Gold) [de] Date 2010-12-31 00:46
Actually I just wonder if the adapter works with the latest Aquarium version :)
Parent - By Felix Kling (Gold) [de] Date 2010-12-31 00:48
Parent - - By ChessMonster (***) Date 2010-12-31 01:29
I own Jiri's book since several months, and a few days ago, I also bought the Hiarcs book. They have a quite different approach to openings, but I must say I like Jiri's book better for tournament chess due to its width. Still I think anybody would need to buy both to remain competitive.

No comments about this cluster idea yet, I think this way seems to be the best for Mr. Rajlich to protect his intellectual investment in chess programming. I mean all these chess programmers need financial motivation to continue what they do, so if mr. Rajlich feels that his talent will be paid better in this business model, he should go for it. I find his motivation quite normal.

I wonder how much Aquarium nodes this thing can crank in 1 hour at a fixed depth. From what I've read in CPU ranking lists, this cluster is roughly 20x faster than a dual Xeon setup? Or more? I am unsure.

If this cluster were to publish a new Jiri's book kind of chess opening book, I would pay 100-200$ for it, but for just 1 hour of analysis, I think only the very best players would rent it, but it might be sufficient to make mr. Rajlich financially fulfilled.
Parent - - By Felix Kling (Gold) [de] Date 2010-12-31 01:38
afaik Jiri used the cluster while working on his book :)
Parent - - By mdraith (**) [de] Date 2010-12-31 01:50
To verify his ideas, I don't think that he did thousands of cluster games to compile the book.
Parent - - By Felix Kling (Gold) [de] Date 2010-12-31 02:08
yes, sure. Manual analysis is still better than monte carlo like stuff.
Parent - - By ChessMonster (***) Date 2010-12-31 02:27
I thought chess was kind of solved after seeing Jiri's book to be honest. I am a FM, and all those lines backed with deep computer analysis, I thought chess was dead. I looked my lines in Hippo, and saw ideas for white that lead to a win. Any opening I looked at , I saw many ideas untried before in human or computer play.

Now, months have passed, I used tons of Aquarium, so my own version of Jiri's book is even stronger than Jiri's original book, but it is only on the few openings I play. I have some novelties in critical lines that fully equalize the Black side, which are neither on Jiri's nor on Hiarc's book.

So if Jiri's book was redone with this new cluster, I would not hesitate to spend another 100-200$ for this new book, because I know it would be improved quite a lot. And I think that would be what most folks are ready to pay for this cluster, not a hourly rate. Hourly rate would be paid by people or governments trying to make your players top in the world, which is still a good and valuable income for the Rybka team, so it makes sense business-wise.

Anyway, I think chess is solved like checkers. Maybe, we will see another 10-20 year of increased human play at top level, but then it won't be as fun anymore, because all will be solved more and more. I don't know, maybe we choose the wrong sport :-/
Parent - - By mdraith (**) [de] Date 2010-12-31 02:32
The Rybka 4 book was compiled for and with the aid of the Rybka cluster. There will be no new cluster book in the near future, I guess.
Parent - - By Felix Kling (Gold) [de] Date 2010-12-31 02:36
Rybka 5 will have a new book, I'm pretty sure about it :)
Parent - - By mdraith (**) [de] Date 2010-12-31 02:42
Sure, because you have to pay extra to get your hands on it. :twisted:
Parent - - By vrboska (**) [de] Date 2010-12-31 11:52
Yes, that's the strategy. No Update, a new Rybka 5 with new book to earn money. :twisted:
Parent - - By Felix Kling (Gold) [de] Date 2010-12-31 14:08
Who said there's no update? ;)
Parent - - By vrboska (**) [de] Date 2010-12-31 14:59
Hey Felix, you told me estemated at April - let's see what is quicker, the update or Rybka 5 ;-)
Parent - - By Arrière Pensée (Gold) Date 2010-12-31 20:02

>Yes, that's the strategy. No Update, a new Rybka 5 with new book to earn money


That is what you first posted-

> let's see what is quicker, the update or Rybka 5 ;-)


Then you change up with this-

Stop trolling!
Parent - By vrboska (**) [de] Date 2011-01-01 13:45
Hi sidserious, first a happy New Year and all the best for 2011!
Perhaps my English isn't good enough to bring details on a point. What I want to say is: I expect no update - although it is often announced, I expact a charged Rybka 5. No trolling - maybe a poor English. I wanted to underline my caustic statement with a smily - sorry if you didn't recognised it. And I hope this forum gives a critical and paying customer the right to say his conviction: what happened with us Rybka 4 customers is a poor customer friendliness!!!
Parent - - By ChessMonster (***) Date 2010-12-31 02:43
Not now, but maybe in a year or two, there might be financial motivation by the Rybka team to publish a new and deeper book. As I said, I would not hesitate to spend money on it. Jiri's book is really good. I quit chess a long time ago, because I had no contact with sufficiently strong players or coaches to improve my game, especially openings, but this Jiri's book really make you very strong in opening, even if you don't consult any GM or SGM.
Parent - - By mdraith (**) [de] Date 2010-12-31 02:51
An honest word from a top player counts more for me than the marketing blabla of some engine freaks. Thanks!
Parent - - By ChessMonster (***) Date 2010-12-31 03:06
1) How do you know they would be honest with their opening advice? I think they keep their golden treasure only to themselves, and a very close circle of friends.

2) Do you know how much their honest opening advice would cost compared to a very good computerized book? The computerized book like that of Jiri is much more cost effective for the average player.

3) Don't call anybody freak, it is impolite and slightly disrespectful.

4) Those SGMs you talk about, they are the ones that use computers most extensively these days altough you might be unaware. Most of them have a supercomputer base at home, although none as strong as the Rybka cluster.

5) I am not marketing anything, I am not a Rybka or Convecta employee, I am just a very happy customer who wants to thank this computer "freak" crowd for fixing my openings for so cheap.
Parent - - By mdraith (**) [de] Date 2010-12-31 03:09
:eek: I was actually talking about you. And sorry for saying engine freaks to engine freaks. For me this has nothing to do with chess, sorry.
Parent - - By ChessMonster (***) Date 2010-12-31 03:35
So you called me top player? I have to thank for the compliment in this case :-)

I quit competitive chess around age 19, (I was rated 2300+ elo at age 15) because I had no good coach and getting GM or SGM advice was too costly 10-15 years ago.

But now, look at this opening books by Jiri and Hiarcs combined. Add in your own Aquarium analysis if you want to improve on their work (which I did)

Your openings will be SGM like... No I am not kidding, it is just I am really grateful for this effort they did. I feel I got awesome advice on my openings within 2 months of studying Jiri's book with Aquarium. That is just invaluable for a tournament player who is not a SGM yet. Even SGMs have holes in their repertoire by the way. I found many holes in Vigiutov's French book or Bologan's books. Do you think these holes are lack of their honest advice? no, it is just the complexity of chess, they can't know everything. Their books are great way to start you to learn an opening I can't deny that, but the width, the depth and the number of ideas for me to test in Aquarium + Jiri's book is so plenty.

Similar move needs to be done in academia that would free us from the dependency of the top universities in the world. I feel they are not sharing their knowledge enough with the rest of the world, but instead try to hold on to their academic status by keeping knowledge to themselves. ( I know totally unrelated topic, but i see many similarities)

Anyway, I'll buy Rybka5 and I'll also buy Rybka 5 opening book, but please make rybka 5 opening book as wide as current Jiri's book, and be generous in sharing cluster knowledge. Jiri's book was quite good, don't lower the standards you've set.
Parent - By mdraith (**) [de] Date 2010-12-31 03:41
Jiri's book was not optimized for Rybka 4, hopefully this will be true for the next Rybka books as well.
Parent - By Felix Kling (Gold) [de] Date 2010-12-31 03:15
he meant you with "top player". I know, that had to be confusing for you ;)
Parent - - By Felix Kling (Gold) [de] Date 2010-12-31 02:34
Well, chess is too complex for a human to remember all the lines. So even if computers solve it at some point, the human games won't be boring :) And engines will still be needed to find the moves that cause humans the most problems.
Parent - - By mdraith (**) [de] Date 2010-12-31 02:39
Chess solved? The 6 men tablebases can't be stored properly because the amount of data is too large. What you suggest is that someday there will be 32 men tablebases. Yeah, go on ranting guys... :lol:
Parent - - By Felix Kling (Gold) [de] Date 2010-12-31 02:44
There is a difference between practically solved and completely solved. I think it's realistic to assume that there will be a solution in sense of having analysed all the main lines to the end.
Parent - - By nebulus (****) [no] Date 2010-12-31 02:49
Then a side line becomes a new main line and you have to start all over again :wink:
Parent - - By ChessMonster (***) Date 2010-12-31 03:14
True :-) But you see , what I mean by solved, there will be fewer openings that people play, because most will be refuted until the endgame. And yes the few openings that people play will be switching sidelines and mainlines. So people will eventually also exhaust those openings. You are underestimating the total collective parallel CPU power at the hands of the chess players in the world. One guy runs a quadcore CPU and publishes his Aquarium results in a tree. Another guy handles a sideline, or improves the line already out there, etc... I feel with a good opensource Aquarium effort, exhausting chess is not a dream.
Parent - - By mdraith (**) [de] Date 2010-12-31 03:19
Then I play 1. a4 and you're out of luck. Do you know how hash tables work? Every position is crunched down to a 64 bit value. You know how many different positions are possible in a chess game? Do you know the error margins of that technique? Do you know how errorprone selective pruning is? I wouldn't trust engine analysis more than how far I can throw a stone.
Parent - - By Felix Kling (Gold) [de] Date 2010-12-31 03:21
let's speak again in 30 years about the subject. 30 years ago from now you would have told me that a human will always be better than a computer because there are so many positions :)
Parent - - By mdraith (**) [de] Date 2010-12-31 03:35
More nodes than atoms in the universe. Maybe the Rybka cluster can handle this? :yell:
Parent - By Razor (****) [gb] Date 2010-12-31 09:40
I believe Felix is correct here - I remember recently seeing a piece written by Dr John Nunn, one of the greatest problem solvers of our time.  It wasn't the article that left a lasting impression, it was the statement in the footer that went something like, "in maths you have to be right, in chess you have to be more right than your opponent"!  :)
Parent - - By Felix Kling (Gold) [de] Date 2010-12-31 03:42
sure, it doesn't have to be a "complete" solution. Just exclude "bad" or "unlikely" positions and you get a much smaller number.
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka Cluster Rental Program
1 2 3 4 5 6 7 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill