CEGT 40/20: starting the tests with Rybka2.3.2a x64 4CPU

[Werner]

Hi all,
we have now started the matches against Rybka2.3.2a 64bit 4CPU.
The first opponents will be:
Zap!ChessZanzibar x64 4CPU
Naum2.2 x64 4CPU
Naum2.2 x64 2CPU
Hiarcs 11.1 4CPU
Hiarcs 11.2 2CPU

The match can online be followed under the link:

http://www.husvankempen.de/nunn/werner

And in our forum and of course here every sunday you can have a look at the results of the other matches.

http://husvankempen.de/nunn/phpBB2/viewforum.php?f=1

[ArmyBridge]

Hi Werner
Very nice!! we will see if cegt get similar result of ccrl test, it seems that Rybka don´t scale very well in 4 cpu

[Uri Blass]

Hi Werner
Very nice!! we will see if cegt get similar result of ccrl test, it seems that Rybka don´t scale very well in 4 cpu

No

We are not going to see it based on this test.

Rybka 2 cpu already had the following results:

54.0 % against Zap!Chess Zanzibar x64 4CPU
69.0 % against Naum 2.2 4 CPU
71.5 % against Naum 2.2 2 CPU
74.2 % against Hiarcs 11.1 4CPU
73.6 % against Hiarcs 11.2 2 CPU

If rybka cannot improve much from 70% to better results then it still does not mean that rybka cannot improve much from 50% to better results.

The only way to test if rybka 4 cpu earn from speed is to give more time to the opponents and test both 4 cpu and 2 cpu.

If the target is to get true rating then it is better to use time control of 40/100 and set rybka strength on 20% and test exactly against the same opponents(except zap that is not uci engine so it cannot be tested under arena).

Rybka will get the same conditions that it get in 40/20 but the opponents will get better conditions.

The result also can be used for Getting rating for
Naum 2.2 4 CPU*5,Naum 2.2 2 CPU*5,Hiarcs 11.1 4CPU*5,Hiarcs 11.2 2CPU*5

Uri

[Werner]

[The only way to test if rybka 4 cpu earn from speed is to give more time to the opponents and test both 4 cpu and 2 cpu.
Uri

Hi Uri,
and from where do you get the strength of this engine mit the twice time e.g. Do you think anybody has the time to make thousand games with such a time control to find it out?
And do you know a GUI which supports different time controls in eng-eng matches?
Using an autoplayer let the engines play with ponder off what does not help very much
We tried to make a better rating for the Rybka 2CPU engines when giving the other engines up to 4 cores. This is a little bite easier but needs also a lot of games to get the rating of all these 4CPU engines.
Perhaps someone with statistical knowledge can say more. I was said not to make matches between engines which have more than 200-250 elo differences.

[ernest]

............
And do you know a GUI which supports different time controls in eng-eng matches?
............

The Chessbase GUI allows in the Engine vs. Engine window, to set for the Engines a "Time increase" of 0 to 200%

[Werner]

............
And do you know a GUI which supports different time controls in eng-eng matches?
............

The Chessbase GUI allows in the Engine vs. Engine window, to set for the Engines a "Time increase" of 0 to 200%

Thanks Ernest,
I didn´t use that before
Just tested: 100 % means double time and +200 % threefold time.

[Uri Blass]

[The only way to test if rybka 4 cpu earn from speed is to give more time to the opponents and test both 4 cpu and 2 cpu.
Uri

Hi Uri,
and from where do you get the strength of this engine mit the twice time e.g. Do you think anybody has the time to make thousand games with such a time control to find it out?
And do you know a GUI which supports different time controls in eng-eng matches?
Using an autoplayer let the engines play with ponder off what does not help very much
We tried to make a better rating for the Rybka 2CPU engines when giving the other engines up to 4 cores. This is a little bite easier but needs also a lot of games to get the rating of all these 4CPU engines.
Perhaps someone with statistical knowledge can say more. I was said not to make matches between engines which have more than 200-250 elo differences.

Arena1.1 allows to set the engine strength to every number between 1% and 100%(of course it can be done for every uci engine but cannot be done for Fritz or Junior or only chessbase programs)

I tested rybka2.3.2a 32 bit (engine strength 10%) against Movei and at time control of 40/40 on PIV 3.19 ghz rybka won 47-33

I expect the result to be worse at slower time control and on time control of 40/10 on A3000(2.01 ghz) that is faster computer than the pIV movei was leading against rybka when I stopped the match.

Uri