Houdini - Cluster Rybka encounter....

[GenoM]

Doc, they do not understand you and they even not trying to. That;s the reason for this big misunderstanding

PS:
haha, napravo sym angelche

[Dr.Wael Deeb]

Doc, they do not understand you and they even not trying to. That;s the reason for this big misunderstanding

PS:
haha, napravo sym angelche

Fact

[M ANSARI]

Houdini is not that brilliant as many think.....

It's just that Rybka is making horrible positional blunders and is getting squashed by Houdini.....

Something is wrong here,I don't know.....

Is it a bad special settings for the match

Is it a bad scaling and not getting the optimal communication with all the 64 processors that Rybka is using

Is it an experimental version of Rybka that is palying

Again,after watching the second game,I am pretty much convinced that Houdini is not that ingenius chess entity as much as there is definitely something wrong with cluster Rybka.......
Dr.D

I agree with you that Cluster Rybka did not play both games verry well (or strong as expected e.g. by me).
Houdini did well in both games. Most peoples interpret too much in this games, thats a little pity. It is just an interesting event, and both engines should get the same respect.

Regards, Clemens keck

I tend to agree. For one thing, there is simply no way the 2-0 actually represents their relative strength in these configurations. Despite having roughly 4x the horsepower, Houdini 3 is probably still the stronger player, though by how much is unclear. I am guessing 20-40 Elo.

The reason is that jumping from 16 cores to 64, is nothing like 2 to 8, and the general loss in parallel efficiency drops badly. So two double ups in speed would normally yield ~80-100 Elo, but here less, and on equal platforms Houdini 3 is already at least 100 Elo stronger than Rybka 4.1. Add to that that this is not actually a 64-core machine, but is a 64-core cluster, and the efficiency is bound to be even worse.

Don't get me wrong, Houdini played very well, and they were fun games, however it would be quite interesting if the owner of the 16-core machine that played, could test Deep Rybka 4.1 to get some comparisons in depth as opposed to the 64-core cluster.

Still, it's reasonable to say that up-to-date Rybka engine on the cluster is not the level of Houdini 3 (both on equal hardware). So, there are small chances we will see soon a Rybka 5 of the strength of Houdini 3 or stronger, and this fact can make news. Also, if parallelization is bad and Rybka engine is actually strong, then again, it shows these clusters generally as not very effective.

Kai

There is a huge difference between a 64 core cluster using memory that is on 8 different connected machine and a 64 core cluster running on shared memory on one motherboard. I am not sure what the 16 core setup H3 was using, but most likely the Rybka 4.1 setup was using a bunch of 8 core or 12 core motherboards. So the strongest single motherboard of the cluster is actually weaker than the H3 stand alone setup. Now if you have followed how the cluster works, it is basically a multi variation version of Rybka 4.1 running at full strength on each of the variations. This is the Master + Slave setup that was used very successfully in Centaur play. I think a 64 core Cluster is only competitive against engines that don't use a motherboard that is stronger than it lowest common denominator motherboard. Against a 16 core H3 running shared memory, I would think that the Cluster would have to be configured with 5 or 6 16core machines. Add to that the fact that H3 is between 70 to 100 ELO stronger than Rybka 4.1 and a competitive Cluster would have to use many more machines.

What I did notice was that the Cluster was using a Rybka evaluation that is at least 3 years old and nothing has been done to improve that. This is antiquated in computer chess terms, and to be honest Rybka 4.1 still has problems stalling, and needs a lot of code cleanup. It really does look like Vas has completely given up Computer Chess and unless there is a Rybka 5 on his computer that he is working on, this looks like this would be it for Rybka.

If Vas would come back and start work on Rybka, I wouldn't be surprised if he could quickly get Rybka competitive again. The almost 300 cores on the cluster would be an incredible tool in improving Rybka's evaluation and they would have H3 and many other strong engines to tune the evaluation. This is how the original Rybka slowly improved and gained on Shredder, and I don't see why it can't happen again. What is certain though is freezing everything and hoping hardware will overcome the evaluation and other shortcomings will not cut it.

[elpapa]

I think a 64 core Cluster is only competitive against engines that don't use a motherboard that is stronger than it lowest common denominator motherboard.

So the cluster is totally meaningless?

[AdminX]

Houdini will fall sooner than later and you will be defending the new king of the hill eventually......
Infinite loop regards,
Dr.D

Houdini can't fall.
It is computerized, it will keep its strength.

I might add it will keep its strengths as well as it's weakness. Same goes for Cluster Rybka.

[Dr.Wael Deeb]

Houdini will fall sooner than later and you will be defending the new king of the hill eventually......
Infinite loop regards,
Dr.D

Houdini can't fall.
It is computerized, it will keep its strength.

I might add it will keep its strengths as well as it's weakness. Same goes for Cluster Rybka.

Correct..........

[Vinvin]

I think a 64 core Cluster is only competitive against engines that don't use a motherboard that is stronger than it lowest common denominator motherboard.

So the cluster is totally meaningless?

"cluster" means "less well interconnected processor than SMP"

[Uri]

I found this article in Wikipedia. Perhaps someone here who is more knowledgable in computer science or has a Ph.D in this area can improve this Wikipedia article:

A computer cluster consists of a set of loosely connected computers that work together so that in many respects they can be viewed as a single system.

The components of a cluster are usually connected to each other through fast local area networks, each node (computer used as a server) running its own instance of an operating system. Computer clusters emerged as a result of convergence of a number of computing trends including the availability of low cost microprocessors, high speed networks, and software for high performance distributed computing.

https://en.wikipedia.org/wiki/Computer_cluster

[S.Taylor]

Houdini will fall sooner than later and you will be defending the new king of the hill eventually......
Infinite loop regards,
Dr.D

Houdini can't fall.
It is computerized, it will keep its strength.

I might add it will keep its strengths as well as it's weakness. Same goes for Cluster Rybka.

I mean, even if another engine overtakes something to become number one, the older number one cannot "fall". It's a strange terminology for computers. It will stay as it is. It will still see the same combinations etc.

[Dr.Wael Deeb]

Houdini will fall sooner than later and you will be defending the new king of the hill eventually......
Infinite loop regards,
Dr.D

Houdini can't fall.
It is computerized, it will keep its strength.

I might add it will keep its strengths as well as it's weakness. Same goes for Cluster Rybka.

I mean, even if another engine overtakes something to become number one, the older number one cannot "fall". It's a strange terminology for computers. It will stay as it is. It will still see the same combinations etc.

The word "fall" indicates that the particular engine is no more at the top of the particular rating list.......

Bearing in mind this definition,it's not a strange terminology Shimon,it's simply a fact......
Dr.D