Stockfish 2.3.1 update available

[Jim Ablett]

http://www.stockfishchess.com/

Stockfish 2.3.1 JA by the Stockfish Team

This update fixes the chess960 bug.

1) For people compiling themselves the Stockfish bench signature is: 5423738
It means that if you run ./stockfish bench then it should show 5423738 nodes searched.
This means that the compile is correct.

2) For people running / using under Android / iPhone with more than one thread,
I'd suggest to set "Use Sleeping Threads" option.
This allows to save power and battery and strength is practically the same.

Window/Linux/Android 32/64 bit
https://dl.dropbox.com/u/5047625/stockfish-231-ja.zip
Mirror:
http://cl.ly/JewK/stockfish-231-ja.zip

Also available from my homepage and official Stockfish website soon.

Jim.

[kasinp]

Jim,

there was talk about including another fix which could be performance related. Did that happen?

Thx,
Peter

[Jim Ablett]

Jim,

there was talk about including another fix which could be performance related. Did that happen?

Thx,
Peter

Hi Peter,

I'll leave Marco or Gary to comment on the specifics, but in the meantime
here you will find all the info on latest code improvements >
https://github.com/mcostalba/Stockfish/commits/master

Jim.

[kasinp]

Jim,

there was talk about including another fix which could be performance related. Did that happen?

Thx,
Peter

Hi Peter,

I'll leave Marco or Gary to comment on the specifics, but in the meantime
here you will find all the info on latest code improvements >
https://github.com/mcostalba/Stockfish/commits/master

Jim.

Appreciate the response, looks like we'll have to wait for a clarification.
Thanks again,
Peter

[Jim Ablett]

Bonus for rook/queen attacking pawns on same rank

Patch and tuning by Gary Linscott from an idea of Ryan Taker.

Double tested by Gary:

Wins: 3390 Losses: 2972 Draws: 11323
LOS: 99.999992%
ELO: 8.213465 +- 99%: 6.746506 95%: 5.124415
Win%: 51.181792 +- 99%: 0.969791 95%: 0.736740

And by me:

After 5612 games 1255 1085 3271 +11 ELO

Jim.

[MM]

http://www.stockfishchess.com/

Stockfish 2.3.1 JA by the Stockfish Team

This update fixes the chess960 bug.

1) For people compiling themselves the Stockfish bench signature is: 5423738
It means that if you run ./stockfish bench then it should show 5423738 nodes searched.
This means that the compile is correct.

2) For people running / using under Android / iPhone with more than one thread,
I'd suggest to set "Use Sleeping Threads" option.
This allows to save power and battery and strength is practically the same.

Window/Linux/Android 32/64 bit
https://dl.dropbox.com/u/5047625/stockfish-231-ja.zip
Mirror:
http://cl.ly/JewK/stockfish-231-ja.zip

Also available from my homepage and official Stockfish website soon.

Jim.

Hi,

2.3.1 JA 64 bit SSE42 stopped at chess960 again in the following (1st position).

qnnbbrkr/pppppppp/8/8/8/8/PPPPPPPP/QNNBBRKR w HFhf - 0 0

It also seems to count less knodes than 2.3 from the classic starting position at 1 core.

FRITZ 12 gui
Intel i7 980x
windows 7

Best Regards

[Ajedrecista]

Hello Maurizio:

http://www.stockfishchess.com/

Stockfish 2.3.1 JA by the Stockfish Team

This update fixes the chess960 bug.

1) For people compiling themselves the Stockfish bench signature is: 5423738
It means that if you run ./stockfish bench then it should show 5423738 nodes searched.
This means that the compile is correct.

2) For people running / using under Android / iPhone with more than one thread,
I'd suggest to set "Use Sleeping Threads" option.
This allows to save power and battery and strength is practically the same.

Window/Linux/Android 32/64 bit
https://dl.dropbox.com/u/5047625/stockfish-231-ja.zip
Mirror:
http://cl.ly/JewK/stockfish-231-ja.zip

Also available from my homepage and official Stockfish website soon.

Jim.

Hi,

2.3.1 JA 64 bit SSE42 stopped at chess960 again in the following (1st position).

qnnbbrkr/pppppppp/8/8/8/8/PPPPPPPP/QNNBBRKR w HFhf - 0 0

It also seems to count less knodes than 2.3 from the classic starting position at 1 core.

FRITZ 12 gui
Intel i7 980x
windows 7

Best Regards

I am an outlier in Chess 960, but your FEN string may be wrong. Arena 2.0.1 reports this position as 'position 9':

Code: Select all

FEN: qnnbbrkr/pppppppp/8/8/8/8/PPPPPPPP/QNNBBRKR w KQkq - 0 1

The problem can be in the castling rights. Is HFhf correct? Or maybe HAha or KQkq? Please take a look at this topic:

FRC castlings questions

I do not know what FEN string is used in Fritz GUI. You can also give a try to '0 1' instead '0 0' at the end of the FEN string.

I am sure that Maurizio needs a much better help than mine.

Regards from Spain.

Ajedrecista.

[gladius]

Bonus for rook/queen attacking pawns on same rank

Patch and tuning by Gary Linscott from an idea of Ryan Taker.

Double tested by Gary:

Wins: 3390 Losses: 2972 Draws: 11323
LOS: 99.999992%
ELO: 8.213465 +- 99%: 6.746506 95%: 5.124415
Win%: 51.181792 +- 99%: 0.969791 95%: 0.736740

And by me:

After 5612 games 1255 1085 3271 +11 ELO

Jim.

Yes, 2.3.1 should be a bit stronger than 2.3. Big thanks to Ryan for the great idea. And Marco of course for all the hard work!

[MM]

Hello Maurizio:

http://www.stockfishchess.com/

Stockfish 2.3.1 JA by the Stockfish Team

This update fixes the chess960 bug.

1) For people compiling themselves the Stockfish bench signature is: 5423738
It means that if you run ./stockfish bench then it should show 5423738 nodes searched.
This means that the compile is correct.

2) For people running / using under Android / iPhone with more than one thread,
I'd suggest to set "Use Sleeping Threads" option.
This allows to save power and battery and strength is practically the same.

Window/Linux/Android 32/64 bit
https://dl.dropbox.com/u/5047625/stockfish-231-ja.zip
Mirror:
http://cl.ly/JewK/stockfish-231-ja.zip

Also available from my homepage and official Stockfish website soon.

Jim.

Hi,

2.3.1 JA 64 bit SSE42 stopped at chess960 again in the following (1st position).

qnnbbrkr/pppppppp/8/8/8/8/PPPPPPPP/QNNBBRKR w HFhf - 0 0

It also seems to count less knodes than 2.3 from the classic starting position at 1 core.

FRITZ 12 gui
Intel i7 980x
windows 7

Best Regards

I am an outlier in Chess 960, but your FEN string may be wrong. Arena 2.0.1 reports this position as 'position 9':

Code: Select all

FEN: qnnbbrkr/pppppppp/8/8/8/8/PPPPPPPP/QNNBBRKR w KQkq - 0 1

The problem can be in the castling rights. Is HFhf correct? Or maybe HAha or KQkq? Please take a look at this topic:

FRC castlings questions

I do not know what FEN string is used in Fritz GUI. You can also give a try to '0 1' instead '0 0' at the end of the FEN string.

I am sure that Maurizio needs a much better help than mine.

Regards from Spain.

Ajedrecista.

Hi,

yes the position is number 9 but i meant that it was the 1st position i tried.

I just copied the fen automatically from fritz, i don't think it is wrong.

By the way i used at least 20 different engines/versions (including several versions of Stockfish) and none stops running in that position.

I believe there's still a bug.


Best Regards

[Ajedrecista]

Hello Peter:

Jim,

there was talk about including another fix which could be performance related. Did that happen?

Thx,
Peter

Both Jim and Gary confirm that there will be a small Elo gain. Details about tests can be found at Pull 23 in GitHub.

I specially take a look on two tests:

Gary's test:

Ah, yes, so, having the condition be >= RANK_5 is important I think. Here are the final results from the rook7th branch, with your optimization.

ELO: 8.22 +- 99%: 5.79 95%: 4.40
LOS: 100.00%
Wins: 4406 Losses: 3837 Draws: 15757 Total: 24000

Using an own tool, I get that the Elo gain is more less +8.24 ± 2.58 Elo with 95% confidence. It means that the worst case is 8.24 - 2.58 = 5.66 Elo of improvement.

Marco's test:

And by me:

After 5612 games 1255 1085 3271 +11 ELO

Using my own tool again, I get more less +10.53 ± 5.87 Elo with 95% confidence. Again, the worst case is 10.53 - 5.87 = 4.66 Elo of improvement. Just a typo: 1255 + 1085 + 3271 = 5611 games and not 5612... but the difference is too small!

As self tests tends to overestimate the Elo gap between two different versions of the same engine, let me say that the worst case can be 3 or 4 Elo of improvement over 2.3 version, but the most probable case will be over 5 and even over 10 if we are lucky... but it is almost a random guess.

Regards from Spain.

Ajedrecista.