Secret engines that beat Rybka 4.

Discussion about chess-playing software (engines, hosts, opening books, platforms, etc...)
Post Reply
Charles
Posts: 71
Joined: Thu Jun 10, 2010 7:41 pm
Real Name: Charles
Contact:

Secret engines that beat Rybka 4.

Post by Charles » Thu Jun 24, 2010 5:15 pm

:mrgreen:

There are some secret settings when applied to Firebird 1.1 that make it much stronger:
the first is what Dr. W. D suggested which is:

PH=64 Pawn=120,knight=bishop=325,rook=600,queen=1170,bish pair=65 Verification Reduction=8

Then there is my CS_6 settings which are:

PH=64, pawnvalue=100,knight=bishop=300,rook=500, queen=900,bishppair=66,Verification Reduction=8

In nunn 20 positions suite +3s per move on Arena 4 cpu ponder=off , results are:

Code: Select all

FireBird_11_in_CS6 - Rybka 4 w32     : 20.5/40 10-9-21 (=====1110===0000=====0=11=1=11=0==0==110)  51%  

FireBird_11_DrWD - Rybka 4 w32 : 21.0/40 12-10-18 (=10101===10=10==10=10===010=0=11=1=1==0=)  53%  
They seem about even but currently running a 20 game gauntlet match using 10 positions and Rybka 4 is losing badly to both -- and interestingly enough I have doubled the time control for this match +6s per move ...

Code: Select all

FireBird_11_DrWD - Rybka 4 w32   : 10.5/19 4-2-13 (1===1=1=0======10==)  55%   
-----------------FireBird_11_in_CS6-----------------
FireBird_11_in_CS6 - Rybka 4 w32 : 11.5/19 6-2-11 (=0=1======1111=10==)  61%   

All tests above 4 cpu p=ff 32 bit. 256Hash

Just 2 more games to go and rybka 4 has already lost both matches.

I know .. not enough games but Dr W D has done his own testing and confirmed much better results than the default.

My theory is that Firebird 1.1 was strong but needed to be tuned , while FB1.2, FB1.3 changes were not necessarily for the better.

User avatar
Dr.Wael Deeb
Posts: 104
Joined: Thu Jun 10, 2010 8:29 pm
Real Name: Dr.Wael Deeb

Re: Secret engines that beat Rybka 4.

Post by Dr.Wael Deeb » Thu Jun 24, 2010 6:06 pm

Thanks charles for the testing and your new settings :D
FireBird 1.1 is definitely the strongest version of the program and these personalities are even stronger....
Looking closely at your settings,it seems like the values you use are what Bobby Fischer has suggested except that he suggest 325 for the bishop,so 100 300 325 500 900....I have the inner feeling that if you change the bishop value in your settings to 325 it will beat both of the settings posted above :mrgreen:
Cheers,
Dr.D

PENSADOR
Posts: 3
Joined: Wed Jun 16, 2010 5:08 pm

Re: Secret engines that beat Rybka 4.

Post by PENSADOR » Sat Jun 26, 2010 12:49 am

64 bit or 32 bit?

Normal compilation, KLO compilation, KLO popcnt?

It varies from machine to machine you know.

User avatar
Dr.Wael Deeb
Posts: 104
Joined: Thu Jun 10, 2010 8:29 pm
Real Name: Dr.Wael Deeb

Re: Secret engines that beat Rybka 4.

Post by Dr.Wael Deeb » Sat Jun 26, 2010 9:59 am

PENSADOR wrote:64 bit or 32 bit?

Normal compilation, KLO compilation, KLO popcnt?

It varies from machine to machine you know.
x64 Intel compilation....
Dr.D

Charles
Posts: 71
Joined: Thu Jun 10, 2010 7:41 pm
Real Name: Charles
Contact:

Re: Secret engines that beat Rybka 4.

Post by Charles » Sat Jun 26, 2010 5:57 pm

Looks like Rybka 4 strikes back this time using mfl modified -- 30 positions suite this time.
Arena tc=+3s hash=256 ph=64 for fire 4 cpu ponder =off 32 bit.
Rybka 4 w32 - FireBird_11_DrWD : 33.5/60 21-14-25 (====0==0=1=0110=1=1010==11==0=1=1=11=1==10=0111=101=10100==0) 56%
Rybka 4 w32 - FireBird_11_in_CS6 : 34.5/60 23-14-23 (1100==1010==1==0=110=010==1=01111==01==01110==101=1==0==1=11) 58%

:!:

Different suites different results -- the reason why I like using test suites vs the same old boring opening book.

Zherkovv
Posts: 36
Joined: Thu Jun 10, 2010 10:12 am

Re: Secret engines that beat Rybka 4.

Post by Zherkovv » Sun Jun 27, 2010 8:46 am

Thanks for the settings.

Post Reply