OpenChess

Independent Computer Chess Discussion Forum
https://open-chess.org/

Cool Iris 10.40

https://open-chess.org/viewtopic.php?t=3678

Page 28 of 30

Re: Cool Iris 10.40

Posted: Thu Mar 14, 2024 2:12 pm
by Peter
Anton101 wrote: Thu Mar 14, 2024 1:48 am Cool Iris 12.10: https://pixeldrain.com/u/LMF7vwvX

Cool Iris 12.10-D3072: https://pixeldrain.com/u/uZw9AqV7
Thanks for reintroducing internal MultiPV
:)

Re: Cool Iris 10.40

Posted: Thu Mar 14, 2024 2:41 pm
by Peter
crikoch wrote: Thu Mar 14, 2024 8:01 am the new vesrion can't solve this :-(

1r3rk1/3bbppp/1qn2P2/p2pP1P1/3P4/2PB1N2/6K1/qNBQ1R2 w - - 0 1
You'r right, that's regression, but it's one that SF dev. 240303 (the last one I downloaded) shares with latest Cool Iris.
SF 16.1 with 30 threads of 16x3.5GHz and 8G hash:

Code: Select all

1r3rk1/3bbppp/1qn2P2/p2pP1P1/3P4/2PB1N2/6K1/qNBQ1R2 w - - 0 1

Analysis by Stockfish 16.1:

1.Sbd2 
  Schwarz hat entscheidenden Vorteil.  
  -+ (-2.90)  Tiefe: 1/4   00:00:01  1764Knoten
...
1.De1 g6 2.Dh4 Tfe8 3.Th1 h5 4.gxh6 
  Schwarz steht deutlich besser.  
  -+ (-2.48)  Tiefe: 7/12   00:00:01  79kN
1.Th1 g6 2.Th6 Lf5 3.Dh1 Lxd3 4.Txh7 Da2+ 5.Sbd2 
  Schwarz hat entscheidenden Vorteil.  
  -+ (-2.81)  Tiefe: 8/12   00:00:01  156kN
...
1.Th1 Dbxb1 2.Txh7 Lh3+ 3.Txh3 Tb2+ 4.Kg3 Dxc1 5.Dxc1 Dxc1 6.Lh7+ Kh8 7.Lf5+ Kg8 
  Die Stellung ist ausgeglichen.  
  = (0.00)  Tiefe: 22/23   00:00:02  29505kN
1.Lxh7+ Kh8 2.Th1 gxf6 3.exf6 Le8 4.Lc2+ Kg8 5.Dd3 
  Weiß steht besser.  
  +/- (1.01)  Tiefe: 23/24   00:00:02  33689kN
...
1.Lxh7+ Kxh7 2.Th1+ Kg8 3.Th8+ Kxh8 4.Dh1+ Kg8 5.g6 Dbb2+ 6.Lxb2 fxg6 7.Sg5 Dxb2+ 8.Kg3 Df2+ 9.Kxf2 Txf6+ 10.exf6 Lxf6 11.Dxd5+ Kh8 12.Sf7+ Kh7 13.Dxd7 Sd8 14.Sxd8 Txd8 15.Dh3+ Kg8 16.Ke3 a4 17.De6+ Kh7 18.Kd3 Lg5 19.De5 Lc1 20.De1 Lh6 21.c4 Tb8 22.c5 a3 23.Sc3 Lf4 24.Dh4+ Lh6 25.Dh1 Kg8 26.Dd5+ Kh8 27.Dg2 Lf4 28.Kc4 g5 29.Sb5 g4 30.Sxa3 g3 31.Sb5 g6 32.Sc3 Kg7 33.Se4 Kf8 34.Sxg3 g5 35.Se2 
  Weiß steht klar auf Gewinn.  
  +- (5.36)  Tiefe: 37/76   00:00:42  1516MN
Tried three times, each one about as fast as that.

SF 240303:

Code: Select all

  
1r3rk1/3bbppp/1qn2P2/p2pP1P1/3P4/2PB1N2/6K1/qNBQ1R2 w - - 0 1

Analysis by Stockfish dev-20240303-b0ac8a4e:

1.fxe7 Sxe7 
  Schwarz steht auf Gewinn.  
  -+ (-3.95)  Tiefe: 1/3   00:00:01  2471Knoten
...
1.Th1 h6 2.De1 Lg4 3.fxe7 Sxe7 4.gxh6 Lf5 5.hxg7 Dg6+ 6.Kf2 Lxd3 7.gxf8D+ Kxf8 8.Sbd2 Dxc3 9.De3 Sf5 10.Df4 Ke7 11.e6 Dg3+ 12.Dxg3 Sxg3 13.Kxg3 Tg8+ 14.Kf2 fxe6 15.Sb1 Lxb1 16.Ld2 Dc2 17.Txb1 Dxb1 18.Lxa5 Dg6 
  Schwarz steht besser.  
  -/+ (-1.16 ++)  Tiefe: 33/58   00:01:57  3607MN
1.Lxh7+ 
  Weiß steht etwas besser.  
  +/= (0.69 ++)  Tiefe: 33/58   00:01:57  3609MN
...
1.Lxh7+ Kxh7 2.Th1+ Kg8 3.Th8+ Kxh8 4.Dh1+ Kg8 5.g6 Dbb2+ 6.Lxb2 Dxb2+ 7.Kg3 fxg6 8.Sg5 Df2+ 9.Kxf2 Txf6+ 10.exf6 Lxf6 11.Dxd5+ Kh8 12.Sf7+ Kh7 13.Dxd7 Tf8 14.Dxc6 Txf7 15.Ke2 Lxd4 16.cxd4 Tf6 17.Dh1+ Kg8 18.Sc3 a4 19.Se4 Te6 20.Kd3 a3 
  Weiß steht auf Gewinn.  
  +- (3.85)  Tiefe: 33/58   00:01:57  3619MN
...
1.Lxh7+ Kxh7 2.Th1+ Kg8 3.Th8+ Kxh8 4.Dh1+ Kg8 5.g6 Dbb2+ 6.Lxb2 Dxb2+ 7.Kg3 fxg6 8.Sg5 Df2+ 9.Kxf2 Txf6+ 10.exf6 Lxf6 11.Dxd5+ Kh8 12.Sf7+ Kh7 13.Dxd7 Sxd4 14.Dh3+ Kg8 15.Dh8+ Kxf7 16.Dxb8 Lh4+ 17.Kf1 Sf5 18.Dc7+ Ke6 19.Db6+ Ke5 20.Dxa5+ Kf4 21.Ke2 Sg3+ 22.Kd3 Sh5 23.c4 Lg3 24.Sd2 Kg4 25.c5 Sf4+ 26.Kc4 g5 27.c6 Sg2 28.Se4 Lf4 29.c7 Lxc7 30.Dxg5+ Kf3 31.Sd2+ Ke2 32.Dxg2+ Ke1 
  Weiß steht klar auf Gewinn.  
  +- (5.62)  Tiefe: 39/81   00:02:45  5336MN
So almost 2 minutes instead of 2 seconds, at a second one trial correct solution was not found even after 3'.
MultiPV can't help here so much neither, because it's a matter of 3.Rh8+ to be found, not only of more than one primary split at first move,

Re: Cool Iris 10.40

Posted: Thu Mar 14, 2024 3:25 pm
by Anton101
Hi Peter!

Sure all the next Cool Iris will have the code "Random".

You are welcome.

Re: Cool Iris 10.40

Posted: Thu Mar 14, 2024 3:27 pm
by Anton101
Anyway I left again two downloads.

Cool Iris 12.10 architecture: 2056

Cool Iris 12.10-D3072 architecture: 3072

Re: Cool Iris 10.40

Posted: Thu Mar 14, 2024 5:45 pm
by OrgZ
:difus_19

Re: Cool Iris 10.40

Posted: Thu Mar 14, 2024 9:38 pm
by Anton101
;)

Re: Cool Iris 10.40

Posted: Thu Mar 14, 2024 9:51 pm
by Cengiz-Baba
Anton101 wrote: Thu Mar 14, 2024 1:48 am Cool Iris 12.10: https://pixeldrain.com/u/LMF7vwvX

Cool Iris 12.10-D3072: https://pixeldrain.com/u/uZw9AqV7

Happy test!
Dear Anton,

My last test is in play now, it's half over. It's a tournament without a book. Something's come to my attention. CI version 12.10 is playing in a very strange style without a book. I didn't realise what changed, but you will understand me when I upload the tournament results and matches to the site. Is there a change in your playing style?

Best regards,
Fringe.

Re: Cool Iris 10.40

Posted: Thu Mar 14, 2024 11:43 pm
by Anton101
Hi!

Default setting for Random MultiPV=5 for solving position test.

For tactical analysis Random MultiPV=2

Best setting for playing vs engines (private test, chess server) Random MultiPV = 1.

Best regards;

Re: Cool Iris 10.40

Posted: Fri Mar 15, 2024 12:26 am
by Cengiz-Baba
Hi Anton,

I was testing at MV=1. The engine has very strange opening preferences. An unbalanced series of defeats occurred between the engines it was a clone of (unbooked tournament). I will post it on the site when the tournament is over. As someone who has 2500 blitz elo in Lichess, I could not understand this imbalance. If there was a player in the tournament and a human, I could say that he changed his style.

Thanks and best regards
Fringe.

Re: Cool Iris 10.40

Posted: Fri Mar 15, 2024 1:51 am
by Anton101
Well, the engine was tested with 250 games, which is the maximum number of games I can allow for time, pc, etc..., between derivatives and the latest SF development, Cool Iris 12.10 D-3072 did not produce any "noise", it did not generate any "bugs". "In any case, it will be very positive if you post the pgn and over the weekend I will start studying and working on the next version.
All times are UTC+01:00
Page 28 of 30

Powered by phpBB® Forum Software © phpBB Limited