CEGT 40/20 update (Juli 15th 2007)

[Werner]

Hi all ,

Our newest rating list and games for download are now online. All links at the end of the posting as usual.

40/20
This week we decided to eliminate both Strelka versions from our list. But we will save the games and see what happens in the future. With the permission of Ryan Benitez we test our 2nd private (beta) engine: Fruit 070327-SF1agg (search focus 1 and playing style aggressiv). The second setting provided by Marc Lacrosse was only tested by me. Fruit 070327-SF1agg will be also listed in our 40/120 list in the future. This is a tribute to Fabien Letouzey who did a lot for our hobby. Hope we will soon see an official update of his engine.

New in our list is Matacz 1.3 - best Polish engine in 6th OPCCC.

Booot 4.13.1
Booot 4.13.1 did hold its good ranking from last week!

Matacz 1.3
Matacz 1.3 has a first rating of 2475 after 245 games. This is about 80 points more than the version 1.08 in our list.

Fruit 070327-SF1agg
The new Fruit version starts with 2808 elos after 188 games in our list. We are all eager what is possible with this engine. Thank you Ryan, we wish you success.

4CPU-engines
With our 4CPU-engines I have made now more than 300 games each. And till today Michael can help me to make some more games .
Here is our first ranking:

Code: Select all

             4CPU    2CPU   diff. 
Zanzibar     2990    2931  +59 
Naum2.1      2918    2840  +78 
Hiarcs11.1   2916    2871  +45 
Loop13.6     2888    2835  +53 
LoopM1-P     2888    2853  +35 
Paderborn    2885    2839  +46 
DShr10       2883    2862  +21 
LoopM1-T     2875    2863  +12 
DFritz10     2853    2893  -40

A big thanks to all testers as usual.

Links:
All-Versions: http://www.husvankempen.de/nunn/40_40%2 ... liste.html
Best-Versions: http://www.husvankempen.de/nunn/40_40%2 ... liste.html
Blitz: http://www.husvankempen.de/nunn/blitz.htm (also with Best-Versions and All-Versions lists)
40/120: http://www.husvankempen.de/nunn/40_120_ ... liste.html
Replay-Zone: http://www.husvankempen.de/nunn/Replay/replay.htm
Elo comparison 40/120 and 40/20: http://www.husvankempen.de/nunn/Replay/ ... arison.htm
Testers with picture and Hardware: http://www.husvankempen.de/nunn/testers/testers.html

Werner Schüle
CEGT-Team

[Werner]

Next week we will have a regular update of our 40/120-list. But just we tested Rybka 2.3.2a 64 2CPU in 15 Matches with 50 games each against the same opponents as already before the other Rybkas.
The result of these games was a nearly unbelievable increase at this level of 50 elos to 3045.

A not official list you can find here at the end of this thread:

http://husvankempen.de/nunn/phpBB2/viewtopic.php?t=784

[Spock]

Yes, CCRL show a 40-50 ELO improvement also on the 2CPU version on both blitz and 40/40, although we don't have your amazing 40/120
It is incredible, but sadly the 4CPU version has not shown the same level of improvement for us.

[Uri Blass]

Next week we will have a regular update of our 40/120-list. But just we tested Rybka 2.3.2a 64 2CPU in 15 Matches with 50 games each against the same opponents as already before the other Rybkas.
The result of these games was a nearly unbelievable increase at this level of 50 elos to 3045.

A not official list you can find here at the end of this thread:

http://husvankempen.de/nunn/phpBB2/viewtopic.php?t=784

I am more surprised by the 2 cpu versus 1 cpu improvement that you have in your list.

I expected 50 elo improvement for doubling the speed.
2 cpu against 1 cpu is less than doubling the speed but I find bigger improvement than 50 elo based on your list(note that I do not know if Fritz10 and Deep Fritz10 are the same engine because I remember that in the past Deep Fritz was better than Fritz with the same number even with single cpu).

13 Spike 1.2 Turin 2CPU : 2829 16 16 1000 46.4 % 2854 19 Spike 1.2 Turin : 2779 17 17 1000 40.8 % 2843 37.7 %

50 elo improvement


6 Deep Fritz 10 2CPU : 2884 17 17 950 56.3 % 2840 38.8 %
15 Fritz 10 : 2803 19 19 850 45.2 % 2837 33.4 %

81 elo improvement

[Werner]

6 Deep Fritz 10 2CPU : 2884 17 17 950 56.3 % 2840 38.8 %
15 Fritz 10 : 2803 19 19 850 45.2 % 2837 33.4 %
81 elo improvement

Hi Uri,
I think these may have several reasons:
- statistical (+-12 points in 40/20 and +-18 in 40/120)
- different engines (should be tested)?
- different behaviour in short or longer time controls
( F10 has elo diff. 30 and DF10 14 if you compare 40/120 and 40/20 results)

- perhaps more reasons but sure not the opponents, because on 40/120 all engines play against the same opponents.

I will also make more games with DF10 on 4CPUs - this engine is sure as strong as 2CPU version...

[rdan1987]

I will also make more games with DF10 on 4CPUs - this engine is sure as strong as 2CPU version...

Nice rating list!!...it would be interesting though if CEGT would test Deep Junior 10.1 4 CPU or Glaurung 2/e5 4 CPU!!

[Werner]

Nice rating list!!...it would be interesting though if CEGT would test Deep Junior 10.1 4 CPU or Glaurung 2/e5 4 CPU!!

Yes, we will include Junior, Glaurung and Rybka...

[M ANSARI]

So it does seem that 2.3.2a x64 is quite a bit stronger than Rybka 2.3 The strongest engine just got even stronger ... hard to believe. I wonder how much stronger Rybka can get. Obviously Rybka still has some weaknesses and can yet be improved, but really we must be reaching some sort of wall here. What I find incredible is that latest Rybka is about 250 ELO points stronger than commercial Fruit. I remember when Fruit was the big thing and at that time I was wondering if it was possible to improve on that. Obviously 250 ELO later ... I guess the answer is yes. Will there be a new revolutionary engine that will gain another 250 ELO's in 2 years ???

[Spock]

So it does seem that 2.3.2a x64 is quite a bit stronger than Rybka 2.3

That's true on 1CPU and 2CPU, but our testing shows that it is not the case on 4CPUs. We show a mere 6 ELO gain from Quad 2.2 x64 to Quad 2.3.2a x64.

I look forward to CEGT running 4CPU tests as well, to see if they get a similar result

[Uri Blass]

So it does seem that 2.3.2a x64 is quite a bit stronger than Rybka 2.3 The strongest engine just got even stronger ... hard to believe. I wonder how much stronger Rybka can get. Obviously Rybka still has some weaknesses and can yet be improved, but really we must be reaching some sort of wall here. What I find incredible is that latest Rybka is about 250 ELO points stronger than commercial Fruit. I remember when Fruit was the big thing and at that time I was wondering if it was possible to improve on that. Obviously 250 ELO later ... I guess the answer is yes. Will there be a new revolutionary engine that will gain another 250 ELO's in 2 years ???

1)This is not correct that rybka is 250 elo stronger than fruit

From the CEGT 120/40 list:
1 Rybka 2.3.2a 64 2CPU : 3026 30 29 344 73.3 % 2851 18 Fruit 2.2 : 2777 18 18 850 42.1 % 2833 40.5 %

You can see that rybka used 2 cpu so this comparison is not fair.
Note also that rybka used 64 bits when fruit used 32 bits.

2)I do not believe that rybka reached a wall and
Rybka clearly has bugs and weaknesses.


Uri