Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka Cluster in Chessbase cloud
1 2 3 4 Previous Next  
Parent - By M ANSARI (*****) [kw] Date 2012-08-06 07:01
I think many just don't realize how much some are willing to spend on a hobby.  Lukas really enjoys working with computers and hardware and one of the best ways to push hardware is via chess software, and if you actually look at the dollar amount spent, it is probably less than a super duper car or even a small yacht or plane.  But you have to realize that it takes tremendous skill and a lot of experimentation to blaze the trail and create a properly working well synchronized bunch of computer hardware to run an OS and then the Rybka Cluster.  I think that fits well with Vas wanting to also build the strongest chess entity possible on the planet.  If you look at even the earliest Rybka's they are all process based engines rather than threaded (I don't know of any other process based engine that is in the top) ... the only conceivable reason to do that is because it would be a good idea once hardware caught up.  Cluster chess software is still in its infancy, but you really cannot compare a normal chess engine to a cluster engine.  There is just so much work involved in getting efficient returns from your hardware as you ramp up in cores, and the simple idea of just running the cores as MP simply fades in returns after 32 cores.  I think there is still a huge amount of improvement to be made in efficiency, and most likely the solution will be in specific modules that are good at certain things.

Another thing to note is have this huge cluster is a tremendous tool in tuning the engine's evaluation.  Vas was the one that revolutionized chess engine tuning via millions of very fast games ... I would expect he has figured out a few other tricks, and would be shocked if Rybka Cluster did not have the most accurate chess evaluation of any chess engine.
Parent - By saurus_ (**) [gb] Date 2012-08-07 10:55
+1
Parent - By tomgdrums (****) Date 2012-08-05 21:52
Dream matchup would be some sort of playoff tournament with mini-matches (like the recent candidates).  Each match consisting of 6 games.  And then blitz for tie breaks if needed. 

With these 8 engines (when available) all on equal cores (say 6-12):

Houdini (I guess wait for H3)
The most recent Rybka cluster software version
Critter 1.6
Komodo 5 MP 
Most recent Ivanhoe
Stockfish
Naum 4.2
Junior 13

The top 8 in CCRL's pure list.

Of course this is a dream tournament that won't happen but I think it would be super fun!!

First pairings by seeding would be:

Houdini vs junior
Critter vs Naum
Stockfish vs Komodo
Rybka vs ivanhoe (the grudge match :)  )
Parent - - By Gaмßito (****) [cr] Date 2012-08-06 10:26
If you can play the dream match between Houdini 3 and Rybka Cluster, please do it. This could give us enough information of how really strong is the Cluster. This will be a valuable information; the cluster strength information still continue hidden to 99% of people and I hope it may be known soon.

Also I recommend you to play the match under equal conditions. It will be even more exciting. That means that Houdini 3 would need to have at least 4 or 5 times more time (you must do the maths) to compensate the difference in hardware. Otherwise the Cluster will have too much advantage.

BTW, what processors use your 12 cores computer? It's overclocked?

Regards,
Gaмßito.
Parent - By Carl Bicknell (*****) [gb] Date 2012-08-06 10:37
it's 12 cores @ 4.22 Ghz
Parent - - By Uly (Gold) [mx] Date 2012-08-06 11:41
The only way to know the real Cluster software strength is to take out the hardware, MP Speedup, Cluster Speedup, and possible inefficiencies out of the formula: Test Houdini 2.0 1CPU against Cluster 1CPU. It should be a good indicator unless the 1CPU of the user is very slow.
Parent - - By Gaмßito (****) [cr] Date 2012-08-06 13:14

>The only way to know the real Cluster software strength is to take out the hardware, MP Speedup, Cluster Speedup, and possible inefficiencies out of the formula: Test Houdini 2.0 1CPU against Cluster 1CPU. It should be a good indicator unless the 1CPU of the user is very slow.


Yes I agree, but what would happen if the Cluster on equal hardware could be less than +40 Elo points stronger than the next Houdini 3? Maybe the business does not collapse but I am sure at least many ''rich'' users will keep away from renting it.

I think the differences should be notorious to survive in this business.

Gaмßito.
Parent - - By Uly (Gold) [mx] Date 2012-08-06 13:18
I consider Houdini 3 vaporware at this point. It could never appear, it could be weaker than Houdini 2.0. So what I care about now is the results of Houdini 2 against the Cluster and how its software compares.

The people salivating on the Cluster may be for a big surprise if it turns out it's mostly hardware.
Parent - - By Banned for Life (Gold) Date 2012-08-06 14:06
The people salivating on the Cluster may be for a big surprise if it turns out it's mostly hardware.

A big surprise? :lol: The Cluster is designed to efficiently use a large numbers of cores on a single position. It is NOT designed for use on a single core. I doubt that Lukas ever tests this way. People salivating over the cluster are doing so because it is the most powerful processing that can be applied to a single position, and by a large margin.
Parent - - By Uly (Gold) [mx] Date 2012-08-07 01:35
Yeah, but what's the value if it's mostly hardware? What if it could be made stronger by getting Rybka out of it and putting Houdini, Critter, Stockfish or Komodo on there? People find evident that Rybka must be the best on there and that it has the strongest software on earth, but it hasn't been tested, and Rybka is 60 elo behind Houdini, practically tied with other 2 engines, and Rybka 4.1 1CPU weaker than Komodo 5 1CPU by 32 elo.

If you believe Cluster software ought to be stronger than other software, I also hope you pray to god, to stay consistent.
Parent - - By Banned for Life (Gold) Date 2012-08-07 04:16
Yeah, but what's the value if it's mostly hardware?

A cluster solution is one where you are assuming a set of networked computer nodes with specialized software running on each node. It would be very surprising if the cluster software ran as efficiently on a shared memory platform as the software that has been optimized for that application. The bottom line is that a cluster solution requires both distributed software, and sufficient hardware to make up for the lower efficiency on a single node. Testing on a single node is anything but a reasonable test.

Taking this a step further, there is only one top engine that is currently being optimized for single core usage, and that of course is Komodo. Once you make the leap to MP with shared memory, certain inefficiencies have to be accepted, so people who are saying that once Komodo goes MP it will be the top engine are possibly deceiving themselves.
Parent - - By Uly (Gold) [mx] Date 2012-08-07 05:49
Agreed, but the people that think Rybka Cluster software is greatly improved from Rybka Public software could be deceiving themselves also: perhaps Vas has focused on improving parallelism performance so much that he has left the core software part unattended.
Parent - - By Banned for Life (Gold) Date 2012-08-07 06:17
I think you are misstating this. Here is what I would concede:

The people that think Rybka 5 UCI software is greatly improved from Rybka 4/4.1 could be deceiving themselves also; Perhaps Vas has focused on improving search parallelism performance for the cluster and hasn't worked on improving shared memory search or evaluation.

For the cluster application, an improvement in search efficiency across multiple nodes might be achieved by compromising the performance on a single node. This is why your original test suggestion wouldn't be valid.
Parent - - By Ray (****) Date 2012-08-07 06:21

> The people that think Rybka 5 UCI software is greatly improved from Rybka 4/4.1 could be deceiving themselves


Well, if Rybka 5 was not significantly stronger than Houdini, I don't think Vas would release it at all. So if there is to be one, we have to assume that it will be. However, Rybka 5 is very much vaporware at the moment. As is Houdini 3 for that matter.
Parent - - By Banned for Life (Gold) Date 2012-08-07 15:53
I think R5 will only be released as the strongest engine, but I don't think it would be more than 20 or 30 Elo better than Houdini.
Parent - - By Carl Bicknell (*****) [gb] Date 2012-08-07 16:27

> I think R5 will only be released as the strongest engine, but I don't think it would be more than 20 or 30 Elo better than Houdini.


This is a really, really big ask though. Houdini 3 is supposed to be 50 elo stronger than Houdini 2. Let's say it's only 30 elo stronger. That would mean Rybka 5 would have to be over 100 elo stronger than Rybka 4.1 AND on top of that Vas wants to save some of his developments for the rental project...
Parent - By Banned for Life (Gold) Date 2012-08-07 16:47
Assuming that Vas has been hard at work since R4 came out, I don't doubt that his latest and greatest is already better than what H3 will achieve. But while H3's new ideas will be revealed within six months of release, it seems Vas is hoping the cloud model will provide an alternative that doesn't require him to share his ideas with other developers. I don't know if this will be a successful approach, but it certainly makes it harder for other developers to ride his coattails.
Parent - - By oudheusa (*****) [de] Date 2012-08-07 21:23
I think a 40 elo improvement for R5 over R4.1 will be enough for people to buy it. I don't expect Vas to give away more than 50 elo improvement per release anymore.
Parent - By Ray (****) Date 2012-08-07 21:26
I think he'd be crazy to release R5, knowing that lots of people (including those who won't admit it) steal his ideas. Perhaps he is waiting on some new technology or distribution method that will prevent this from happening.
Parent - By Carl Bicknell (*****) [gb] Date 2012-08-07 21:35

> I think a 40 elo improvement for R5 over R4.1 will be enough for people to buy it. I don't expect Vas to give away more than 50 elo improvement per release anymore.


Yes that sounds reasonable.
Parent - - By Gaмßito (****) [cr] Date 2012-08-07 16:31

> I think R5 will only be released as the strongest engine, but I don't think it would be more than 20 or 30 Elo better than Houdini.


If you meant 20 or 30 Elo better than Houdini 2.0c, then Vas needs to hurry to get the first place, before Houdini 3 comes. As you know, Houdini 3 is supposed to be +50 Elo points stronger than Houdini 2.0c.

Regards,
Gaмßito.
Parent - By Banned for Life (Gold) Date 2012-08-07 16:43
The problem with releases is that once you make them, your ideas are out there and available to anyone willing to extract them via RE. Richard seems to be particularly adept at this process. Vas' advantage is that he is a very sharp guy and can pick up good ideas from other sources and add them to his own to come up with something stronger. I continue to believe that the main reason for Vas not coming out with a newer version of Rybka is that this concept also works in reverse, and he sees the process as self defeating.

I am not sure what kind of strength improvement will come with H3. Certainly Robert has more credibility than some of the lesser engines that are frequently touting 100 Elo improvements, yet remain 100s of Elo behind the leaders. I think it's dangerous to quote the 50 Elo improvement though, as there is no evidence to support this claim. (While the superiority of the Rybka cluster is documented through the tournaments that it played in, there is way too much uncertainty to pin down the magnitude of the improvement).
Parent - - By Uly (Gold) [mx] Date 2012-08-07 06:33

> Here is what I would concede:


Would you concede to this?:

The people that think Rybka 5 UCI software is greatly improved from Rybka 4/4.1, even if Vas threw into it the best he had, could be deceiving themselves also; Perhaps Vas has focused on improving search parallelism performance for the cluster and hasn't worked on improving shared memory search or evaluation.

This is saying that Rybka UCI development could have been greatly damaged by Cluster development, and there's even the possibility that Rybka 4 wasn't a lobotomized version of Cluster Rybka software but that Vas made the best he could with it and yet ended dropping the ball when compared to the great improvement of Rybka 3.

I only have 4 cores, I don't care how Rybka performs in more of them, and I'm not going to rent the Cluster. For people like me, iff this is true, it's a great shame and I'd be wishing the Rybka Cluster was a project that never existed. I think the majority of people aren't going to rent the Cluster, and those would want to know how Cluster Rybka software would run on their machines, that's why a test is in order, and if single CPU software has been weakened because Vas focused on parallelism on many cores, it is important to know, because it'll be the software that will be MP escalated to use the resources of the user's machine.
Parent - By Banned for Life (Gold) Date 2012-08-07 16:02
I agree that Vas might be concentrating on distributed engines, which would mean less improvement for shared memory MP engine performance (and also SP engine performance). Of course Vas has the advantage of not being number one and being able to look at other developers ideas without reciprocation.

The Cluster is only one manifestation of the distributed engine model. The cloud based model is another. It's much too early to say that this won't become the dominant model for all software applications, including chess.

I don't think that SP tests are relevant to anyone not using engines in SP mode (e.g. IdEA users). An engine can be great at SP and not be able to perform MP at all (e.g. Komodo). The distributed model is a further step in the direction of additional parallelism. It's too early to tell if Vas is an early adapter, or tilting at windmills.
Parent - - By AWRIST (****) Date 2012-08-10 12:35
Uly, you are very funny. Are you only a tester or also a player? If the latter you will quickly see how strong it is. Because the RC is for strong players and it is a nice surprise because without the UCI R5 it's difficult for others to present their improvements. If you are more a weaker player RC should be your choice even more because then you beat them all on playchess server. Please dont fear a Komodo or some such. If someone doesnt play chess at all and wanted only to go out on a promenade, THEN RC is sine qua non to get the nicest girls...:cool:
Parent - By Banned for Life (Gold) Date 2012-08-10 15:56
Rolf,

The Rybka Cluster is certainly a significant accomplishment for Vas and Lukas, but I'm pretty sure you would get nicer girls if you spent the time and money on a Ferrari! :lol:
Parent - By Uly (Gold) [mx] Date 2012-08-11 07:00

> If you are more a weaker player RC should be your choice even more because then you beat them all on playchess server.


I'm not interested in playchess server. I play correspondence chess and very much doubt that the money it costs is worth it.

93% of people had this view for one reason or another in 2010:

http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?pid=298710

Only 16 people on the entire forum said they would rent it.

I guess it's time for a new poll.
Parent - - By Lukas Cimiotti (Bronze) [de] Date 2012-08-07 08:30
Basically you are right.
The cluster mainly consists of 2 versions of Rybka. One for the master with added capabilities to control the slaves. And one for the slaves with added communication capabilities to get orders from the master. It's no problem to run the slave like any ordinary UCI engine. It is much stronger than Rybka 4.1.
Parent - - By Uly (Gold) [mx] Date 2012-08-07 08:56

> It's no problem to run the slave like any ordinary UCI engine. It is much stronger than Rybka 4.1.


But, how much? Why hasn't it been tested? Why is it left to the users to have to bid ducats for it to test it? Why neither you nor Vas care about publishing results of this supposed much higher strength?
Parent - By Gaмßito (****) [cr] Date 2012-08-07 09:08

>Why neither you nor Vas care about publishing results of this supposed much higher strength?


By logic, perhaps is because it is not as strong as many people think. Otherwise, I think the results would be published.

Gaмßito.
Parent - - By Carl Bicknell (*****) [gb] Date 2012-08-07 10:46

> But, how much? Why hasn't it been tested? Why is it left to the users to have to bid ducats for it to test it? Why neither you nor Vas care about publishing results of this supposed much higher strength?


As much of a Rybka Cluster fan as I am, Uly makes an excellent point here.

The Cluster costs money to rent and it costs money to run. I wonder if the appeal of renting would be higher if its elo was known to some extent.
Parent - By Mark (****) [us] Date 2012-08-07 11:13

>  I wonder if the appeal of renting would be higher if its elo was known to some extent.


My guess is that the appeal of renting would be lower if the true elo were known, unless the difference was huge, like 250 elo or more.  And you can bet that if there was that much difference, there would be tons of published results available.
Parent - - By Ray (****) Date 2012-08-07 12:16

> But, how much? Why hasn't it been tested? Why is it left to the users to have to bid ducats for it to test it? Why neither you nor Vas care about publishing results of this supposed much higher strength?


I think it is completely irrelevant. You rent 64 cores, not 1. The strength on 1 core vs Rybka 4.1 is just meaningless.
Let's say it is the same as 4.1. Would that stop people renting the cluser ? Of course not. 64 or the 200+ or so cores that are available, it is the strongest chess entity in the world. Nothing more to be said.

When (if) UCI Rybka 5 is released, the strength vs 4.1 is of course relevant then. And only then.
Parent - - By tomgdrums (****) Date 2012-08-07 12:32

>> But, how much? Why hasn't it been tested? Why is it left to the users to have to bid ducats for it to test it? Why neither you nor Vas care about publishing results of this supposed much higher strength?


> I think it is completely irrelevant. You rent 64 cores, not 1. The strength on 1 core vs Rybka 4.1 is just meaningless.
> Let's say it is the same as 4.1. Would that stop people renting the cluser ? Of course not. 64 or the 200+ or so cores that are available, it is the strongest chess entity in the world. Nothing more to be said.


It may be the strongest chess entity on the planet. That has not been proven.

But even if it is the strongest (and it probably is) Uly's statement is incredibly relevant because they are asking money for rental time.  So the consumer needs the all the relevant data (ie: strength, elo and maybe even some games) so that he or she can decide if it is worth THEIR money to rent cluster time.

Personally, for me to even consider renting time with the cluster (even on the Fritz cloud, which you can get outbid which is weird!) the cluster would have to be phenomenally stronger to make it worth my money.  And even then it does not turn me on THAT much.  That is of course a personal decision.  And everyone needs to decide for themselves how they are going to spend their money and chess time.

But if they (Vas and Lukas) are asking for consumers to spend money, they should give their potential customers all the relevant data!  Uly is correct!
Parent - - By Gaмßito (****) [cr] Date 2012-08-07 13:15

> But if they (Vas and Lukas) are asking for consumers to spend money, they should give their potential customers all the relevant data!  Uly is correct!


100% correct yes. And once the results are known, definitely this will help people to choose the best decision: pay or avoid the rent of it.

Regards,
Gaмßito.
Parent - - By Barnard (Bronze) Date 2012-08-07 14:49
Gambito,dont you think that if they tell the true about the rybka development and the cluster strength,maybe the marginally people that rent it will stop doing that?

and from the pure point of the selfish to have a bussines and take benefit,is too much better remain silent about those questions?

p.s.you have a good proof with Richard Lang and his chessgenius program,were he strictly forbiden to run the engine in tournaments,to avoid people knowing the real strength,and continue with his near-dead bussines...do you think that if the marginally people that bought his engine,had know the true strength of chessgenius,they had bought it?

just my 2 cents

regards
Parent - - By Gaмßito (****) [cr] Date 2012-08-07 16:13

> Gambito,dont you think that if they tell the true about the rybka development and the cluster strength,maybe the marginally people that rent it will stop doing that?


Well amigo, it depends on what will be the truth. If the Cluster performance is as expected then it would be very good for Vas and company to make and provide good propaganda. Otherwise (what seems most obvious) it will be bad and I think people will think twice before renting.

>and from the pure point of the selfish to have a bussines and take benefit,is too much better remain silent about those questions?


Yes, it seems more beneficial and logical hide all when the results are not as expected. But if the Cluster performance is great, remain silent seems to me quite absurd. Again, why hide something so good?

> p.s.you have a good proof with Richard Lang and his chessgenius program,were he strictly forbiden to run the engine in tournaments,to avoid people knowing the real strength,and continue with his near-dead bussines...do you think that if the marginally people that bought his engine,had know the true strength of chessgenius,they had bought it?


It depends of how much you like the program, but also the prices were totally different. I can buy a good processor and also other things with the money I need to rent the Cluster now.

BTW, I was a great fan of Chess Genius, Barnard. :smile: I really liked too much the solid human game style this program had, (it was great) and although versions from 5 to 7 showed almost no significant improvements, this did not prevent me to purchase version 7. When we love something, well, you know, we are like that (unless the prices can be exorbitant, of course).

Regards,
Gaмßito.
Parent - By Barnard (Bronze) Date 2012-08-07 16:42

>Again, why hide something so good?


right...the only reason to hide it,is simply:because it isnt as good as it they tell,and showing the truth,can prevent the bussines and people playing for it

>BTW, I was a great fan of Chess Genius, Barnard. I really liked too much the solid human game style this program had, (it was great) and although versions from 5 to 7 showed almost no significant improvements, this did not prevent me to purchase version 7. When we love something, well, you know, we are like that (unless the prices can be exorbitant, of course).


me also.the first time i bought chessgenius,was chessgenius 3 on diskette,and it cost me,in 1996,120 euros of that year!
from that version,the improvement was ridiculous,and from 5 to 7,more than ridiculous

the program died in 1996;at that point,was strong (compared to other programs of that year),but after that,other programs grew up,and chessgenius remain a kid

from what i read and spoke with Richard,the program rely a lot in knowledge,and it scales very poorly in stronger machines

Richard is making his livings from a program that died on 1996
Parent - By BigBen (****) Date 2012-08-08 08:43

>you have a good proof with Richard Lang and his chessgenius program,were he strictly forbiden to run the engine in tournaments,to avoid people knowing the real strength,and >continue with his near-dead bussines...do you think that if the marginally people that bought his engine,had know the true strength of chessgenius,they had bought it?


Do you realise just how wrong that statement is? Lang does not forbid the use of the engine in tournaments (hence 10 world titles) he does not want the engine in rating lists (citation is needed over this point as I assume you have been on WIKI) but it is simple to test the engine run your own match, CG does not scale well on todays modern hardware but it did get some 2795 rating on a 100Mhz machine many years ago but that can be taken with a pinch of salt, it is prob more like 2500 compared to new engines,I wonder how modern engines would run on that hardware?? ...... Yes today chess genius 7 cannot compete with todays modern engines but the development of that engine stopped long ago. In the early days CG and HIARCS were the big boys and for whatever reason Richard got out of the rat race after all he only won about 10 world championships with the hardware that was available at the time and he sold plenty of copies of his program.

Maybe in fact he was ahead of his time by getting into the handheld devices which are powerful when compared to what was used before 2000 ... My HTC phone has a dual core 1Ghz cpu to RUN droidfish etc I am not sure if CG ever run on anything more than a 100Mhz machine.

Is his business near dead then you would have to ask him I have CG for android but thats me but then again I also have shredder which is quite weak when compared to droidfish :lol:

I think elswhere you mention he still sells eeproms for the fantastic tabletop mephistos, what you have to realise that is still a popular market with lets say connoisseurs. If I remember quite a few hang out on the HIARCS forums and if something breaks you need replacements, in fact there is a chap there porting Rybka etc so that it run on these tabletop machines (revelation) http://www.phoenixcs.nl/index.php?option=com_content&view=article&id=135&Itemid=72&lang=en

Regards
Parent - - By Ray (****) Date 2012-08-07 17:56
The relevant data is the strength of the cluster on 64 or 200 cores not 1.
Parent - - By Barnard (Bronze) Date 2012-08-07 18:58
im sure that if you can optimize Crafty to play on 1.048.576 processor,even Crafty will destroy the rybka cluster

what i want to say is rybka rely only on the cluster to win anything;since Houdini appeared,and even more today,rybka will be in serious troubles to win any single tournament,running on equal hardware and the same conditions

and if you take a look to the past,you will notice that a lot of wons for rybka,came from the cluster,not from the engine itself

is the same if i tell you:

ey,lets do a race between you and me...since we never told any limit,you wear comfortable clothes,and then i appear,like rybka appeared,but me mounted on a ferrari,and i tell you:ey,lets start the race!

that means that im better than you running?or it means i want to use every single option that i can to assure the win on our race?

for me,it only means that im really unsure to win against you running,and having a certain fear to losse,i want assure my domination by other methods(me with a car,rybka with a cluster)
Parent - - By Banned for Life (Gold) Date 2012-08-07 19:08
im sure that if you can optimize Crafty to play on 1.048.576 processor,even Crafty will destroy the rybka cluster

This is wrong. Crafty on 2^20 processors will be terrible. Scaling to even 100 processors requires a different manner of handling parallelism.

since Houdini appeared,and even more today,rybka will be in serious troubles to win any single tournament,running on equal hardware and the same conditions

If a frog had wings, he wouldn't bump his ass when he jumped. The Rybka cluster is the world's strongest chess engine, and by a good margin. Of course you can handicap it to make it play weaker by not supplying it with power, or not allowing it to run on a large number of cores, or any of 100 other methods.

that means that im better than you running?or it means i want to use every single option that i can to assure the win on our race?

This shows the Ferrari is much faster than you are with any pair of shoes, just as the Rybka cluster is much stronger than Houdini on any hardware that Houdini will run on.
Parent - - By Barnard (Bronze) Date 2012-08-07 19:37

>if you can optimize Crafty to play on 1.048.576 processor


Alan,i mean what i said:if you can...im sure you cant,was only an hipotesis that even a weak program,if is able to run in enough cores,will defeat the rybka cluster

i wouldnt handdicap rybka,since rybka is rybka,and the cluster is the cluster

if rybka was never released an engine,and from it first aparition was running on the cluster,and only on the cluster,you can tell i want to handdicap rybka,since rybka is embedded only on the cluster,but since rybka was released first as uci en engine,and only years later,the cluster was run,your assumption that im trying to handdicap rybka,is simply wrong

what i want say is clear:

Houdini,on the same hardware,with same conditions,will defeat rybka by scoring at least 58 % of the total points,and near 62 % with Houdini 3,and that only means that rybka is too much weaker than Houdini...simply and clear

so if you need to run an engine on 296 cores to asure the win in a tournament,it only shows the lack of self confidence you have to win in equal conditions

of course,that is legal,but is the same than taking anabolizants in some competitions/countries:is legal,but not fair play
Parent - - By Banned for Life (Gold) Date 2012-08-07 19:53
Alan,i mean what i said:if you can...im sure you cant,was only an hipotesis that even a weak program,if is able to run in enough cores,will defeat the rybka cluster

You can't though, so it's just like saying that if my mother had wings, she'd be an airplane!

Houdini,on the same hardware,with same conditions,will defeat rybka by scoring at least 58 % of the total points,and near 62 % with Houdini 3,and that only means that rybka is too much weaker than Houdini...simply and clear

I don't believe this is true. I suspect that the latest Rybka on similar single computer hardware would be better than the latest Houdini. Of course we won't actually see this unless someone organizes a match between the two with a large enough prize fund to get both to show up.

so if you need to run an engine on 296 cores to asure the win in a tournament,it only shows the lack of self confidence you have to win in equal conditions

The Rybka cluster is much better than Houdini on the fastest machine it will run on. This doesn't mean that the latest Rybka non-cluster wouldn't be better than the latest Houdini.

of course,that is legal,but is the same than taking anabolizants in some competitions/countries:is legal,but not fair play

It's perfectly fair play. Everybody knows what it is, it has been used by some of the world's strongest players, it has won every tournament and freestyle event it has been entered in, and it is clearly the strongest chess playing entity in the world. There is absolutely no justification in saying that it is OK to run on an expensive dual socket Xeon machine, but not on an expensive cluster. The difference is only some additional money for the hardware, and a lot of work to develop software that will support parallelism at a much higher level. Anybody with a fat wallet can do the former. Only Vas has been able to really do the latter.
Parent - - By Barnard (Bronze) Date 2012-08-07 20:13

>I don't believe this is true. I suspect that the latest Rybka


what latest rybka?we only have proof that rybka 4.1 exists,and also we have proofs that Houdini 2.0c exists,and is about 70 elo points stronger than rybka...so please,can you give me any proof that that hipotetical rybka exist?

please,tangible proofs,like rybka 4.1 exists,and Houdini 2.0c exists,and all of us can check that it isnt just a fiction that they invent to continue the bussines of the cluster and also continue the aura of rybka being the top 1 engine

>The Rybka cluster is much better than Houdini on the fastest machine it will run on. This doesn't mean that the latest Rybka non-cluster wouldn't be better than the latest Houdini.


semantics...
and im gonna tell you more:i dislike Houdart,but if i were millionaire,i will build a 1024 cluster for Houdini,will contract Houdart to make a good implementation able to take benefit of those cores,just to destroy once and for all the aura that the wounded animal smell

>It's perfectly fair play. Everybody knows what it is, it has been used by some of the world's strongest players, it has won every tournament and freestyle event it has been entered in, and it is clearly the strongest chess playing entity in the world. There is absolutely no justification in saying that it is OK to run on an expensive dual socket Xeon machine, but not on an expensive cluster. The difference is only some additional money for the hardware, and a lot of work to develop software that will support parallelism at a much higher level. Anybody with a fat wallet can do the former. Only Vas has been able to really do the latter


Alan,read Freud and you will notice what he thinks about people that need to have the biggest 'x' just to prove he is stronger,because,with his own and alone aptitudes,isnt able to do it(engine---ownself aptitudes;cluster----''i must be sure to run in a very stronger machine because i have fear to losse in equal conditions'')

>Only Vas has been able to really do the latter.


wrong;i remember johnie (or whatever it is called) running on 800 cores cluster...now,if you add the elo difference from johnie to Houdini to the guy who developed the paralelism of johnie,you will be able to have a too much more stronger cluster than rybka
Parent - - By Banned for Life (Gold) Date 2012-08-07 20:53
what latest rybka?we only have proof that rybka 4.1 exists,and also we have proofs that Houdini 2.0c exists,and is about 70 elo points stronger than rybka...so please,can you give me any proof that that hipotetical rybka exist?

please,tangible proofs,like rybka 4.1 exists,and Houdini 2.0c exists,and all of us can check that it isnt just a fiction that they invent to continue the bussines of the cluster and also continue the aura of rybka being the top 1 engine


Vas has no desire to prove to you that a later version of Rybka exists, or that the Rybka cluster is X Elo better than Houdini. This is his call. If you wish to deny the obvious because there is no proof, you are entitled to do so, but educated people accept things all the time without proof. In fact, the overwhelming majority of the things that you know have never been proven to you.

and im gonna tell you more:i dislike Houdart,but if i were millionaire,i will build a 1024 cluster for Houdini,will contract Houdart to make a good implementation able to take benefit of those cores,just to destroy once and for all the aura that the wounded animal smell

Once again, the cost of the computers in the Rybka cluster is substantial, but the major cost is the time it took Vas to put the software together. And unlike the UCI versions of Rybka, the cluster version software is secure, so Robert would have to start from scratch. I think it would take him at least two years to come up with a competitive product, and even then I would bet on Vas.

read Freud and you will notice what he thinks about people that need to have the biggest 'x' just to prove he is stronger

All great competitors take advantage of every opportunity to win. As Vince Lombardy said: "Show me a good loser, and I'll show you a loser". In my business, I take advantage of all opportunities to put the best product out. If my competitors aren't keeping up, all the better. Other developers could have spent years of their life developing a cluster so that they had the world's strongest chess playing entity, but they didn't. Instead, they took the easy way out and proclaimed that Vas must have cheated. In reality, Vas is far and away the most innovative guy in the chess development business, and his accusers are the ones who are out of ideas.

wrong;i remember johnie (or whatever it is called) running on 800 cores cluster...now,if you add the elo difference from johnie to Houdini to the guy who developed the paralelism of johnie,you will be able to have a too much more stronger cluster than rybka

Not wrong. The Johnie cluster sucked. It's not just a matter of buying the hardware and burning the electricity. Only the Rybka cluster is generating extra elo from running on hundreds of cores distributed over a dozen computers. Robert doesn't have cluster software today, and it will take a major investment of his time to develop it. Vas has nothing to worry about on this front.
Parent - - By Barnard (Bronze) Date 2012-08-07 22:12
well,we are not going to a common point between your ideas and my ideas,so i think is sterile fight,and even more when the other person,you,have too much better knowledge than me,and i only can speak with 'ideas',and you with ideas plus knowledge

im gonna say just one thing:proof that that better rybka is available to vas:0 and since im enough smart to be an atheist,i dont trust in anything without proofs;blessed people who can trust without proofs,and not only that,also pay for ''smoke''
Parent - - By Banned for Life (Gold) Date 2012-08-07 22:53
There is certainly some indirect proof of the strength of the Rybka cluster. Remember the Mondial freestyle events? Vas and Jirry competed in one of the divisions. The prize fund was pretty good. Jirry came in first using the Rybka Cluster software running on Lukas' hardware. Vas came in second with the Rybka Cluster software running on his own cluster.
Parent - - By Barnard (Bronze) Date 2012-08-07 23:12

>Jirry came in first using the Rybka Cluster software running on Lukas' hardware. Vas came in second with the Rybka Cluster software running on his own cluster


that seems a joke,but having in mind that is right,it can only make laugh loud :smile:

just one question:vas has his own cluster,different from Lukas' cluster?
Parent - - By Banned for Life (Gold) Date 2012-08-07 23:21
Vas has a set of machines he can put together to make a small cluster. I don't think it is comparable to Lukas' cluster in either quality of computers, or quantity.
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka Cluster in Chessbase cloud
1 2 3 4 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill