H3 vs Cluster Rybka

[kinderchocolate]

After the match, who would still rent the cluster?

[Graham Banks]

If Houdini wins the third game we might have to wonder whether the Rybka cluster could even draw a single game...

Looking at the play in the first two games, I think that Rybka might be lucky if it manages to salvage a couple of draws, but I'd be surprised if it won a single game.
Just a gut feeling of course, as one shouldn't make a definitive judgement based on just two games.

[Modern Times]

After the match, who would still rent the cluster?

People who don't have access to a 16 core Xeon box ?

[S.Taylor]

After the match, who would still rent the cluster?

People who don't have access to a 16 core Xeon box ?

I imagine this may deal a terrible blow.

Only someone who is desperate for extremely quick analysis or preparation, and has no access to own strong computer, might consider renting.
Mystique might not be there anymore.

[kinderchocolate]

After the match, who would still rent the cluster?

People who don't have access to a 16 core Xeon box ?

I doubt. The rental is quite expensive. If I can buy a machine better than the cluster with a normal PC price, I wouldn't even click into the rental page. If the cluster isn't the strongest in the world, there is no point.

[Modern Times]

After the match, who would still rent the cluster?

People who don't have access to a 16 core Xeon box ?

I doubt. The rental is quite expensive. If I can buy a machine better than the cluster with a normal PC price, I wouldn't even click into the rental page. If the cluster isn't the strongest in the world, there is no point.

Since when is a 16 core Xeon "normal PC price" ??? They are hugely expensive.

In a way, these matches are completely pointless. If you can't afford to rent the cluster, then you can't afford to buy and run a 16-core Xeon.

Much better would have been a true "normal" (but higher spec) PC - say a single socket 6-core core Intel 3930K or similar - running Houdini 3. If THAT machine could consistently beat the cluster then I agree with you, the cluster is dead. The only exception might be if, on 296 cores, it is world-beater and you have a few million in the bank.

[Houdini]

Since when is a 16 core Xeon "normal PC price" ??? They are hugely expensive.

In a way, these matches are completely pointless. If you can't afford to rent the cluster, then you can't afford to buy and run a 16-core Xeon.

Much better would have been a true "normal" (but higher spec) PC - say a single socket 6-core core Intel 3930K or similar - running Houdini 3. If THAT machine could consistently beat the cluster then I agree with you, the cluster is dead. The only exception might be if, on 296 cores, it is world-beater and you have a few million in the bank.

Clemens' expensive 16-core dual Xeon runs at about 30,000 kN/s.
To obtain the same Elo strength with 6 cores you'd need about 23,500 kN/s.
That's not a lot higher than the typical 18,000 to 20,000 kN/s reached by an overclocked i7-3930K. The $500 6-core CPU delivers 80% of the performance of the $4000 16-core dual CPU.

Robert

[Lion]

I have seen some 3930k at over 21'500 kn/s

Are you sure that a 30'000 kn/s 16 cores is about = to an 6 cores at 21'500 kn/s ?

If its the case, spending so much money for (23.5/21.5) only 9.3% faster cpu is not worth it..... imo

rgds

[Modern Times]

Clemens' expensive 16-core dual Xeon runs at about 30,000 kN/s.
To obtain the same Elo strength with 6 cores you'd need about 23,500 kN/s.
That's not a lot higher than the typical 18,000 to 20,000 kN/s reached by an overclocked i7-3930K. The $500 6-core CPU delivers 80% of the performance of the $4000 16-core dual CPU.

Robert

In which case, that is what should have been used for the match. We can speculate how much better the 16-core is over the 6-core, but speculating inconclusively. Better to have used the overclocked 6-core in the first place. Much better promotion for Houdini as well...

[S.Taylor]

I have seen some 3930k at over 21'500 kn/s

Are you sure that a 30'000 kn/s 16 cores is about = to an 6 cores at 21'500 kn/s ?

If its the case, spending so much money for (23.5/21.5) only 9.3% faster cpu is not worth it..... imo

rgds

I don't see what's bad with i7-3770 either. It's only 4 cores, at 3.4 Mhz each, and can either be overclocked or not, and much cheaper and it's the latest... for high end users (Ivy Bridge).
I do not overclock mine (at the momment), and the whole computer tower stays almost completely cold, day and night, AND it's not such a slow timing is it?