H3 vs Cluster Rybka

Discussion of computer chess matches and engine tournaments.

Moderator: Ras

S.Taylor
Posts: 8514
Joined: Thu Mar 09, 2006 3:25 am
Location: Jerusalem Israel

Re: H3 vs Cluster Rybka

Post by S.Taylor »

Vinvin wrote:Speed :
Rybka cluster 64 cores : 104 Mn/sec
Houdini 3 16*3.4 Ghz : 34 Mn/s

Why use only the 64 core cluster? isn't the rental Rybka on a stronger cluster also?

And hasn't Vas been upgrading his Rybka 4.1 to make it stronger than 4.1 was?

So the Rybka is on 4 times the amount of power than Houdini is? (or not even that?)
Modern Times
Posts: 3799
Joined: Thu Jun 07, 2012 11:02 pm

Re: H3 vs Cluster Rybka

Post by Modern Times »

S.Taylor wrote: Why use only the 64 core cluster? isn't the rental Rybka on a stronger cluster also?
Money.
Lavir
Posts: 263
Joined: Sun Oct 28, 2012 11:45 am

Re: H3 vs Cluster Rybka

Post by Lavir »

S.Taylor wrote: Why use only the 64 core cluster? isn't the rental Rybka on a stronger cluster also?
Will you sponsorize the full cluster with YOUR own money? Do you know how much money it takes to rent the full cluster?

Why if you are so keen on trying to find excuses on why Rybka is obviously better even in front of evidence of the contrary (for now) don't you personally use all your savings so that when you will be bankrupt (much probably) but without the result you expect in return then maybe you will learn the hard way?
S.Taylor wrote: And hasn't Vas been upgrading his Rybka 4.1 to make it stronger than 4.1 was?
I suppose the people who did rent the cluster "downgraded" Rybka to the lower version to make it play worse, naturally.
S.Taylor wrote: So the Rybka is on 4 times the amount of power than Houdini is? (or not even that?)
You are right. 4 times is too little.

Oh wait, wasn't it Rybka the best scaling program on slow time controls without need of ANY hardware help (in yours and Maurizio's mind, at last), and wasn't the Rybka program in the cluster "much better" that Rybka 4.1 anyway?
Suj
Posts: 60
Joined: Thu Dec 24, 2009 7:40 am

Re: H3 vs Cluster Rybka

Post by Suj »

rybka 64 cores on the cloud was giving between 83mnps-85mnps.not sure how that has increased to 103mnps.

If played without book this should be even lesser?

Any way to check exact number of cores?
Vinvin
Posts: 5312
Joined: Thu Mar 09, 2006 9:40 am
Full name: Vincent Lejeune

Re: H3 vs Cluster Rybka

Post by Vinvin »

Suj wrote:rybka 64 cores on the cloud was giving between 83mnps-85mnps.not sure how that has increased to 103mnps.

If played without book this should be even lesser?

Any way to check exact number of cores?
May be we got the faster processors :-)
But that's a terrible advertising for Rybka cluster ...
Suj
Posts: 60
Joined: Thu Dec 24, 2009 7:40 am

Re: H3 vs Cluster Rybka

Post by Suj »

faster e5 xeons instead of 5600 xeons or lesser number of client nodes which inturn gives lesser overheads for greater speed or different cluster settings on speed and queuing and splits or lastly more cores running which might not be the case.
User avatar
Laskos
Posts: 10948
Joined: Wed Jul 26, 2006 10:21 pm
Full name: Kai Laskos

Re: H3 vs Cluster Rybka

Post by Laskos »

The low number statistics is exciting. Playing with the standard deviation of the sample (and not with sample standard deviation), from this 1:0 result I can say with 84% confidence that Houdini 3 16 core is not weaker by more than 100 Elo points than Rybka Cluster 64. That would imply, taking for hardware advantage, that the present day software Rybka used by the cluster has only 16% to be stronger than Houdini 3 software.

Looking at the game, Houdini 3 outplayed Rybka on all chapters. Eagerly awaiting for the second game.
gotogo
Posts: 92
Joined: Tue Jan 26, 2010 5:03 am

Re: H3 vs Cluster Rybka

Post by gotogo »

The Rybka cluster is suppose to represent a level of chess that can not be reached by a high end PC.

Now if the 64 cluster represents a higher level of analysis or anything because it is offered as a cheaper alternative to the 128 and so on it should not sacrifice any strength being a cheaper alternative.

If the 64 cluster can not beat Houdini it should be dropped. Further testing should move to the 128 cluster.

What size is the chessbase cluster offered in the new package and what program is on it? What good will it be if we have the best software avalible with Houdini?

If you ask me the cluster program should be upgraded or dropped...
Maybe Robert could make a cluster program and put Rybka to sleep.
Rybka is like trying to beat an old champion racing horse to beat new fresh young studs, it just isn't going to happen. There is a new young stud in the game and it's Houdini.
S.Taylor
Posts: 8514
Joined: Thu Mar 09, 2006 3:25 am
Location: Jerusalem Israel

Re: H3 vs Cluster Rybka

Post by S.Taylor »

Lavir wrote:
S.Taylor wrote: Oh wait, wasn't it Rybka the best scaling program on slow time controls without need of ANY hardware help (in yours and Maurizio's mind, at last), and wasn't the Rybka program in the cluster "much better" that Rybka 4.1 anyway?
No! None of these things in your post were true, in my mind. (and i was hoping i was right, that not).

I'm very excited to see Houdini winning, even more.
User avatar
Leto
Posts: 2139
Joined: Thu May 04, 2006 3:40 am
Location: Dune

Re: H3 vs Cluster Rybka

Post by Leto »

Nice win by Houdini in game 1. Excited to see the rest of the games.