Rybka Cluster is now available at Chessbase. For a start I set up one cluster with 64 cores.
Cool 
I hope someone will rent and have a showdown in playchess engine room with 12/16 cores houdini :)
Why don't you do it? For the beginning I set a minimum price of 0.0 ducats per minute.
I'm using Fritz13 gui and the beta version of cloud is no longer working, I need to update for DF13. I would be enjoying watching if it will really happen.
>
> I hope someone will rent and have a showdown in playchess engine room with 12/16 cores houdini :)
your wish is my command....
I thought this would take a while? Did you suddenly make progress? This is FANTASTIC news btw !!
p.s what's with the node-count? Is this just an error in the chessbase speed table? (It reads about 3 MN/s)
I think the node count is OK - it's MN, not KN/s.
> I think the node count is OK - it's MN, not KN/s.
It wasn't that that was confusing me. It's Rybka's really low node count! (on my 12 core xeon I get 30 MN/s with Houdini 2)
I guess that's the typical node/s uselessness for engine comparisons :)
And Rybka's weird node counting was meant to fix that
maybe they un-faked the node count to make think people the cluster is stronger than it really is,because 3 Mn/s is out of rybka in 64 cores
you can simply take the kn/s from 1 core running rybka,and now guess how you can obtain 3 Mn/s in 64 cores
you can simply take the kn/s from 1 core running rybka,and now guess how you can obtain 3 Mn/s in 64 cores
the node count has changed from the old cluster.
The 200-core cluster at 18205 kn/s! compared to 64 cores gives 82-85mnps.
The 200-core cluster at 18205 kn/s! compared to 64 cores gives 82-85mnps.
> the node count has changed from the old cluster.
Yes I think they've been adjusted for cosmetic reasons: on the Fritz Cloud people might see the Rybka Cluster producing the same NPS as Houdini on a laptop & they might assume it wasn't really a powerful machine at all.
What is interesting is if 64 cores on the present cluster really gives 85mnps what does the 300 cores or 200 cores rybka cluster give?
If the 200 cores rybka gives more than 200 mnps it really means the cluster algo for rybka is really really good.
More than 1000 nps per core on a cluster is unbeliveably good just like a single core chess engine.
If the 200 cores rybka gives more than 200 mnps it really means the cluster algo for rybka is really really good.
More than 1000 nps per core on a cluster is unbeliveably good just like a single core chess engine.
The cluster's node count doesn't follow the rules valid for Rybka 4. In fact the master's node count is multiplied by the number of computers in the cluster. The idea behind this was to make visible if some of the computers failed to work. Of course this only works for a small cluster. But there's no good reason to put much work in the cluster's node count algorithm as there is only one cluster it will run on
Lukas thanks for the above post.
Does this take into effect the losses due to overheads or is it just an estimate?
How many nodes constitute the 64 core rybka cluster on the cloud?
Does this take into effect the losses due to overheads or is it just an estimate?
How many nodes constitute the 64 core rybka cluster on the cloud?
>Does this take into effect the losses due to overheads or is it just an estimate?
No, it's just an estimate.
The 64 core cluster currently consists of 5 computers: 1 * 16 core E5-2680 and 4 * 12 core Westmere (X5690, 2x X5680, X5670). But that might change.
OK admit it: this isn't just more cores is it? The software has been improved as well. It's eating my testsuite like a shark
do you think that is possible,since you made public you are testing the cluster against other engines solving your testsuite,that they stored the solutions to your testsuite for the case you will try again with the same/almost the same testsuite?
> do you think that is possible,since you made public you are testing the cluster against other engines solving your testsuite,that they stored the solutions to your testsuite for the case you will try again with the same testsuite?
Ha! Well it's possible. I doubt Lukas would do that though. Soon I will have a new testsuite which will require a fresh start.
EDIT: I'm watching this in real time as the cluster solves on the screen. It's all genuine.
well,think like Machiavelo,the end justify the means...
he has a rent progrram to earn money,and if someone running Houdini with 12 cores can overpass the cluster,no one will rent it
you made a testsuite,and with 12 cores,you had betters results running Houdini,than rybka on 40 cores
and apart from that,if you keep in mind that Houdini 3 will be around 50 elo points stronger than Houdini 2,and that means the same than doubling the cores,you will be able to have the same elo you can have today with 12 cores,with a 6 cores when Houdini 3 will be released
and a lot of people have/can afford a 6 core system
you get my point,im right?
well,do you know what Jury (the guy behind Strelka) did to play better Strelka?he stored a pre-analized moves inside Strelka,so Strelka can use it and save time...in other words,Strelka was able to play better moves in less time
he has a rent progrram to earn money,and if someone running Houdini with 12 cores can overpass the cluster,no one will rent it
you made a testsuite,and with 12 cores,you had betters results running Houdini,than rybka on 40 cores
and apart from that,if you keep in mind that Houdini 3 will be around 50 elo points stronger than Houdini 2,and that means the same than doubling the cores,you will be able to have the same elo you can have today with 12 cores,with a 6 cores when Houdini 3 will be released
and a lot of people have/can afford a 6 core system
you get my point,im right?
>EDIT: I'm watching this in real time as the cluster solves on the screen. It's all genuine.
well,do you know what Jury (the guy behind Strelka) did to play better Strelka?he stored a pre-analized moves inside Strelka,so Strelka can use it and save time...in other words,Strelka was able to play better moves in less time
For the Houdini vs Rybka Cluster match I'll wait until Houdini 3 comes out. I can promise you this: on 64 cores RC would crush Houdini 2 on my 12 core. But when Houdini 3 comes out I'll run a match.
As for the testsuite...the Clusters results were impressive, but I couldn't finish testing and it was failing on a few positions at the end. Someone outbid me and was willing to pay over 10 ducats a minute
As for the testsuite...the Clusters results were impressive, but I couldn't finish testing and it was failing on a few positions at the end. Someone outbid me and was willing to pay over 10 ducats a minute
rybka running on 64 cores isnt more than about 30-40 elo points stronger than Houdini running on 12 cores (Houdini is about 70 elo points stronger than rybka,and that is almost the same than an efective doubling 2 times the cpu's)
and if you add the 50 elo points (if Houdart is not lying us) to that elo,you will have that a 12 cores,will be the same strength,or 10 elo points better than the cluster at 64 cores
apart from that,you can overcloock the 12 cores,and if you add,for example,a 50 % more of speed,that means 35 elo points added for the 12 cores running Houdini,and in that case,it will be almost the same strength to the 64 cores RUNNING HOUDINI 2,so if you add the more points that Houdini 3 will have,im more than sure that a overclocked 12 cores running Houdini 3,will surpass rybka 4 on 64 cores
we can bet what you want
and if you add the 50 elo points (if Houdart is not lying us) to that elo,you will have that a 12 cores,will be the same strength,or 10 elo points better than the cluster at 64 cores
apart from that,you can overcloock the 12 cores,and if you add,for example,a 50 % more of speed,that means 35 elo points added for the 12 cores running Houdini,and in that case,it will be almost the same strength to the 64 cores RUNNING HOUDINI 2,so if you add the more points that Houdini 3 will have,im more than sure that a overclocked 12 cores running Houdini 3,will surpass rybka 4 on 64 cores
we can bet what you want
> rybka running on 64 cores isnt more than about 30-40 elo points stronger than Houdini running on 12 cores (Houdini is about 70 elo points stronger than rybka,and that is almost the same than an efective doubling 2 times the cpu's)
>
But here's the issue: it's NOT Rybka 4 on 64 cores. It's the Rybka Cluster. It's better software.
But anyway: look, when Houdini 3 comes out I'll run a match - fair deal?
ok,fair deal when Houdini 3 will be released
I suspect a 64 node cluster will crush H3 if it arrives in September.
What would also be interesting is Rybka Cluster on 12 cores vs Houdini on 12 cores.
you wont see that,for a very explicit reason:
Houdini will destroy rybka(at least Houdini 3,that will be near 120 elo points stronger),and that isnt a good propaganda for them
they play only for the sure win:a 190 cluster (or whatever it was,but more than 100 cores)against 6 cores for Critter,and even with that,Critter drew the game
Houdini will destroy rybka(at least Houdini 3,that will be near 120 elo points stronger),and that isnt a good propaganda for them
they play only for the sure win:a 190 cluster (or whatever it was,but more than 100 cores)against 6 cores for Critter,and even with that,Critter drew the game
> Houdini will destroy rybka(at least Houdini 3,that will be near 120 elo points stronger),and that isnt a good propaganda for them
You do have faith in Houdini than the cluster hey.
I doubt if Houdini 3 will match the Rybka cluster. Don't you think that Vas has improved Rybka?
well,before i give you an answer,we must agree about some terms,so i will be able to give you an honest answer:
1)if i have faith in Houdini like uci program compared to the cluster?that is a nonsense,you can compare apples and cats,and since Houdini hasnt a cluster,i dont get your point
2)if you mean comparing the enginnes,Houdini 3 running in how many cores,and the same for rybka cluster?
3)if i think that vas made improvements to rybka,compared to what?from rybka 3 to rybka 4?from 4.1 to the actual moment?for an uci engine,or just to the software of the cluster?
4)improved,in what way?making it stronger?adding new analisis features for the cluster?for the uci?
regards
1)if i have faith in Houdini like uci program compared to the cluster?that is a nonsense,you can compare apples and cats,and since Houdini hasnt a cluster,i dont get your point
2)if you mean comparing the enginnes,Houdini 3 running in how many cores,and the same for rybka cluster?
3)if i think that vas made improvements to rybka,compared to what?from rybka 3 to rybka 4?from 4.1 to the actual moment?for an uci engine,or just to the software of the cluster?
4)improved,in what way?making it stronger?adding new analisis features for the cluster?for the uci?
regards
> What would also be interesting is Rybka Cluster on 12 cores vs Houdini on 12 cores.
Yes. Or 1 core vs 1 core.
It would be nice if the Rybka team gave an estimate of improvement over Rybka 4.1, it seems a shame to have to hire the machine just to test its strength.
The Rybka Cluster on one core? What's next? Rybka Cluster on a smart phone? When you're looking for the strongest chess entity on the planet, you use whatever hardware you can muster and support. Vas and Lukas have spent a lot of time working on stuff not suitable for SMP. Why should they handicap their best efforts to compete?
What's next? Rybka Cluster on a smart phone? When you're looking for the strongest chess entity on the planet, you use whatever hardware you can muster and support. Vas and Lukas have spent a lot of time working on stuff not suitable for SMP. Why should they handicap their best efforts to compete?
Yes I agree wit you. It's just that an equal match would give the software improvement estimate.
The problem at the moment is that to work out the approx elo of the Cluster we have to rent it. It would be nice to know a rough level of improvement before renting.
The problem at the moment is that to work out the approx elo of the Cluster we have to rent it. It would be nice to know a rough level of improvement before renting.
I guess that with the cluster available on the CB cloud, people will get a reasonable idea of how strong it is. It's kind of a tough problem, because of it's need to play reasonably long TC games, and the expense of running it for the large number of games it would take to come up with reliable numbers.
>Vas and Lukas have spent a lot of time working on stuff not suitable for SMP
i think that here,who spent a lot more money and efforts,was Lukas
im pretty sure that vas only supported to Lukas to build the cluster,with the 'thanks',not with money...or at least Lukas pay the 9o % of the cluster
vas isnt the type of guy who gives money,is the type of guy who wants money
i dont agree in a lot of things with Lukas,but i must admit now,that the effort he made building and paying for the cluster,is inmense...and also im sure without the cluster,rybka had been dead now
Robert is a smart guy, but I suspect he is a distant second to Vas when it comes to chess software development.
The Rybka cluster is the strongest chess entity on the planet, and it is sitting in Lukas' basement. Lukas was putting together state of the art hardware before the cluster came about, and now it has something to chew on. Vas' contribution is of course in the cluster development arena.
The Rybka cluster is the strongest chess entity on the planet, and it is sitting in Lukas' basement. Lukas was putting together state of the art hardware before the cluster came about, and now it has something to chew on. Vas' contribution is of course in the cluster development arena.
in other words:
Lukas paid all the money,and vas only 'put' his handsome face...typical in guys that never gives money,and only want receive it
i dont think that:im sure Robert is very good at software development,but he at least have a minimum of ethics and dont try to find a guy that pay for the large amount of money that the cluster cost,like vas did
in Spanish,we have a good word to describe that behaviour:''caradura''
Lukas paid all the money,and vas only 'put' his handsome face...typical in guys that never gives money,and only want receive it
>Robert is a smart guy, but I suspect he is a distant second to Vas when it comes to chess software development.
i dont think that:im sure Robert is very good at software development,but he at least have a minimum of ethics and dont try to find a guy that pay for the large amount of money that the cluster cost,like vas did
in Spanish,we have a good word to describe that behaviour:''caradura''
Lukas is an MD. His time is valuable. Putting together the world's most powerful chess playing hardware is a hobby of his. If he can make money from it, more power to him. This is a really difficult thing to do because there just isn't any money in chess, but Lukas is a very smart guy, and certainly understood this from the beginning.
Vas is a chess engine developer. I'm pretty sure he is head and shoulders above the rest of the developers. That's why some of the other developers don't like him, especially some of the ones that his algorithms have relegated to 'has been' status.
My dealings with Vas have never left me questioning his ethics. Of course he is a tough competitor, and will take advantage of any opportunities. But to the best of my knowledge, nobody has left Team Rybka because they felt mistreated, and this is the best indication that your thesis is wrong.
Vas is a chess engine developer. I'm pretty sure he is head and shoulders above the rest of the developers. That's why some of the other developers don't like him, especially some of the ones that his algorithms have relegated to 'has been' status.
My dealings with Vas have never left me questioning his ethics. Of course he is a tough competitor, and will take advantage of any opportunities. But to the best of my knowledge, nobody has left Team Rybka because they felt mistreated, and this is the best indication that your thesis is wrong.
i dont think my tesis is wrong:
i agree that Lukas is a very smart person,and i also agree that build the cluster is a hobby for Lukas...
and from that points,i can deduct that Lukas isnt a person who do anything for money,and im sure since he made the cluster for a hobby,he isnt interested in the marginal money that the cluster will make
what im saying is that Lukas put all the money,all the efforts building the cluster,and vas only his 'face',like:im the developer of rybka
what i want to say,or i tried to say,is that Lukas put a lot of more efforts (and of course,money) in the cluster,rather than vas
i agree that Lukas is a very smart person,and i also agree that build the cluster is a hobby for Lukas...
and from that points,i can deduct that Lukas isnt a person who do anything for money,and im sure since he made the cluster for a hobby,he isnt interested in the marginal money that the cluster will make
what im saying is that Lukas put all the money,all the efforts building the cluster,and vas only his 'face',like:im the developer of rybka
what i want to say,or i tried to say,is that Lukas put a lot of more efforts (and of course,money) in the cluster,rather than vas
Remembering that the cluster is a job for Vas and a hobbly for Lukas, the most likely scenario is that Lukas has made the lions share of the hardware investment, but that Vas has a lot more time invested in the cluster software. You seem to be overlooking the difficulties associated with developing the cluster software. If it were easy, lots of people would have good stuff. But there are actually only a few clusters that are any good at all, and the Rybka Cluster is far and away the best of the bunch. This wouldn't have happened without Vas, nor would it have happened without Lukas. It's always good when the people on a team have clearly defined responsibilities, like in this case. This isn't a weakness, it's a strength!
you dont get my point,but is because i cant explain myself better
in sintesis,and to avoid future missunderstanding,what i mean is that vas made all for the money,and Lukas for a hobby,not for the money,but the difference is that im sure vas never paid any single euro for the cluster,and Lukas paid all the costs,including electricity bill
in sintesis,and to avoid future missunderstanding,what i mean is that vas made all for the money,and Lukas for a hobby,not for the money,but the difference is that im sure vas never paid any single euro for the cluster,and Lukas paid all the costs,including electricity bill
I think you are making your points clearly, and I do understand them. But I'm not sure you have a lot of experience with how this type of business is likely to work.
It's a good bet that Lukas has purchased all of the hardware that is located at his residence, and also pays for the electricity (at Germany's exorbitant rates). When cluster time is sold, the income is split between the two. The split formula is whatever is mutually agreed to between Vas and Lukas and isn't any of our business.
As far as income goes, I would be very surprised if Lukas weren't making a lot more money than Vas. Of course this would be from his medical practice, not from the cluster. I would guess that the cluster is operating in the red, but once again, it's a hobby for Lukas. I'm sure he would be happy to make money from it if possible, or at least to pay some of the expenses.
It's a good bet that Lukas has purchased all of the hardware that is located at his residence, and also pays for the electricity (at Germany's exorbitant rates). When cluster time is sold, the income is split between the two. The split formula is whatever is mutually agreed to between Vas and Lukas and isn't any of our business.
As far as income goes, I would be very surprised if Lukas weren't making a lot more money than Vas. Of course this would be from his medical practice, not from the cluster. I would guess that the cluster is operating in the red, but once again, it's a hobby for Lukas. I'm sure he would be happy to make money from it if possible, or at least to pay some of the expenses.
>I'm not sure you have a lot of experience with how this type of business is likely to work.
my experience is near 0
>It's a good bet that Lukas has purchased all of the hardware that is located at his residence, and also pays for the electricity (at Germany's exorbitant rates).
totally agree
>When cluster time is sold, the income is split between the two. The split formula is whatever is mutually agreed to between Vas and Lukas and isn't any of our business
also agree
>As far as income goes, I would be very surprised if Lukas weren't making a lot more money than Vas. Of course this would be from his medical practice, not from the cluster
also agree;from my thoughts,the cluster is only a hobby for Lukas,and im pretty sure he never build it thinking to earn money to make a living,since he doesnt need it
>it's a hobby for Lukas. I'm sure he would be happy to make money from it if possible, or at least to pay some of the expenses.
also agree...
now what i disagree:
for me,to be a good and impartial distribution of the money generated by the cluster,and having in mind that a)Lukas paid all the cluster b)Lukas pays the maintenance c)Lukas pays the electricity bills and updates,i think the most equitative distribution of the money,is 99,99 % for Lukas,and 0,01 % for vas
and im more than secure,because im sure that when rybka was exploding,he alented Lukas to build it,but when the chess,like now,have only a marginal benefits,will be Lukas who will hold with all the expenses,not vas
i think the most equitative distribution of the money,is 99,99 % for Lukas,and 0,01 % for vas
Ah, OK. Maybe next time you can be Lukas' agent. In reality, I don't think Vas would have written the cluster code on that basis. In any event, as a Laissez Faire capitalist, I never even think about this kind of thing since it is up to the parties to come to a mutually acceptable agreement.
Ah, OK. Maybe next time you can be Lukas' agent.
In reality, I don't think Vas would have written the cluster code on that basis. In any event, as a Laissez Faire capitalist, I never even think about this kind of thing since it is up to the parties to come to a mutually acceptable agreement.
>I never even think about this kind of thing since it is up to the parties to come to a mutually acceptable agreement
well,if you contact a hooker(or other service,like a plumber),and knowing that person needs a lot the money,you give only 1/3 part of the money deserved,is also a mutually agreement,right?
you are not the plumber,)of course,not the hooker
),and also you arent who contact him...but it isnt a just agreement,even if you arent involved,im right?...that is my pointLukas paid all,pay for the updates,pay for the failures,pay for the electricity bill,etc,and vas only stam his face on the product
that ISNT just,at least for me...something totally different is that they both agree,but that doesnt make the agreement more just
Lukas paid all,pay for the updates,pay for the failures,pay for the electricity bill,etc,and vas only stam his face on the product
Lukas put the cluster together, which required substantial capital as well as many skills. He also wrote a lot of scripts to make the cluster useful. Vas wasn't just lollygagging around though. He wrote the chess playing software. I suspect this was more difficult than having his face stamped on the box!
I have to admit, I just don't understand the "is it just" element of your argument...
Lukas put the cluster together, which required substantial capital as well as many skills. He also wrote a lot of scripts to make the cluster useful. Vas wasn't just lollygagging around though. He wrote the chess playing software. I suspect this was more difficult than having his face stamped on the box!
I have to admit, I just don't understand the "is it just" element of your argument...
>Lukas put the cluster together, which required substantial capital as well as many skills. He also wrote a lot of scripts to make the cluster useful. Vas wasn't just lollygagging around though. He wrote the chess playing software. I suspect this was more difficult than having his face stamped on the box!
well,semantics
>I have to admit, I just don't understand the "is it just" element of your argument...
i said to you im unable to explain in english what im thinking in Spanish,at least at level of the same expressions that i will use in Spanish
> I would guess that the cluster is operating in the red, but once again, it's a hobby for Lukas
Well I'm sure any loss is a tax deduction. It would be a business, not a hobby
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill