Subject: Re: Chess System Tal 2 released Sat Jul 01, 2023 7:42 pm
mwyoung wrote:
Admin wrote:
From the README -
For Power Users Windows and some UI's can become unstable if threads are set too high, use at your own risk setoption name RelaxThreadLimit value N (set yourself as a power user)
Thanks,
It looks like 4 CPU is best on my computer. That SUCKS!
I do not have that option.
this option is to be used when you have engine console access. i.e. Arena, running from exe, or cutechess
mwyoung
Posts : 880 Join date : 2020-11-25 Location : USA
Subject: Re: Chess System Tal 2 released Sat Jul 01, 2023 7:44 pm
texium wrote:
mwyoung wrote:
Damir Desevac wrote:
Who cares about knodes, important is the playing style.. In case you have not noticed CST2 is using 138 MB nnue.. That would explain the lower nodes on your machine..
Hey Mr. Snark, I only care about if CST2 is running correctly in testing.
What I noticed is not 138 MB nnue.
In this picture you're clearly showing Rebel 14.1 Tal. Not CSTal NNUE 2
already stated that. I pictured the wrong file.
But the issue remains. 3.2 Mnps @16CPU.
CST2 will have to be tested against other chess computers at 4 CPU max.
Chris Whittington
Posts : 1254 Join date : 2020-11-17 Location : France
Subject: Re: Chess System Tal 2 released Sat Jul 01, 2023 7:52 pm
two ways to be a power user:
1. put "poweruser" into the command line
2. UCI command: setoption name RelaxThreadLimit value 1
then you can set threads to be anything up to cores.
Dio likes this post
mwyoung
Posts : 880 Join date : 2020-11-25 Location : USA
Subject: Re: Chess System Tal 2 released Sat Jul 01, 2023 8:08 pm
Chris Whittington wrote:
two ways to be a power user:
1. put "poweruser" into the command line
2. UCI command: setoption name RelaxThreadLimit value 1
then you can set threads to be anything up to cores.
Thanks you are amazing. I will do this for the next test.
Admin likes this post
Mclane
Posts : 3022 Join date : 2020-11-17 Age : 57 Location : United States of Europe, Germany, Ruhr area
Subject: Re: Chess System Tal 2 released Sat Jul 01, 2023 8:58 pm
Ok, now cstal is indeed 8x faster on on the 16 core machine, So my above results were made with 1 core !!!! Impressive.
Admin, mwyoung and Damir Desevac like this post
mwyoung
Posts : 880 Join date : 2020-11-25 Location : USA
Subject: Re: Chess System Tal 2 released Sat Jul 01, 2023 9:05 pm
Mclane wrote:
Ok, now cstal is indeed 8x faster on on the 16 core machine, So my above results were made with 1 core !!!! Impressive.
Yep I like it. CST2 is losing to Stockfish 16 by a bit in my test, but SF 16 is using 16 CPU cores, to CST2 1 or 2 CPU cores at 3.2Mnps only.
mwyoung
Posts : 880 Join date : 2020-11-25 Location : USA
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 12:36 am
C.W. has resurrected my favorite chess player of all time!
Mclane and Eelco like this post
Admin Admin
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 12:51 am
Mclane, TheSelfImprover and mwyoung like this post
texium
Posts : 125 Join date : 2022-07-19
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 4:54 am
Fascinating game where CSTal crushes leela in a sicilian with the black pieces when white leaves book winning. Sacrifices expected. First move out of book is move 14 [pgn] [Event "CSTal vs Leela T1"] [Site "Lucas Chess R 2.08a"] [Date "2023.07.01"] [White "Lc0 v0.30.0"] [Black "CSTal-NN-Strong"] [Result "0-1"] [WhiteElo "3700"] [BlackElo "3700"] [ECO "B78"] [TimeControl "180+2"] [PlyCount "144"] [Opening "Sicilian Defence: Dragon Variation, Yugoslav Attack, Old Line"] [Termination "Mate"]
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 6:22 am
Uses 4 cores okay for me.
texium
Posts : 125 Join date : 2022-07-19
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 8:57 am
do you plan to update it further for elo and style? I noticed that Rebel and CSTal have high draw rates on engine lists which are typically characteristic of engine who have hit a plateau. Ethereal and Berserk were like that before some improvements and still have as high draw rates, like almost 80%. Dragon 3.2 has a slightly lower but high draw rate despite being #2 and so close to SF which only has a 65%
Chris Whittington
Posts : 1254 Join date : 2020-11-17 Location : France
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 12:36 pm
I'm quite disappointed with this version and draw rates. You can expect an upgrade which deals with the opening book plus the cores/threads issues, and with a more aggressive NNUE. Couple of weeks maybe.
Mclane, mwyoung, Nezhman and Ipmanchess like this post
pohl4711
Posts : 159 Join date : 2022-03-01 Location : Berlin
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 1:47 pm
Chris Whittington wrote:
I'm quite disappointed with this version and draw rates. You can expect an upgrade which deals with the opening book plus the cores/threads issues, and with a more aggressive NNUE. Couple of weeks maybe.
As I mentioned before, playing aggressive is not only about playing sacs. This is why my EAS-Tool looks for short wins and bad draws and the average length of all won games of an engine (and adds points for all these things). And here we see, what is still not good in CSTal (EAS/normal):
For CSTal EAS: short wins percent-value is 11.2%, which is a really bad value for an engine, claiming an aggressive playing style. Also the average length of all won games is way too high: 82 moves... (76 moves is the overall average value in my SPCC-Ratinglist and other engines, which play aggressive, have values below 76 (Danasah 9: 70 moves, Komodo 14 aggressive 68 moves (!!!))). And the percentage of bad draws for CSTal EAS is 22.6%, which is too high. Most aggressive engines have bad draw percentages of 17% or below... (what are bad draws for my EAS-Tool? : Bad draws are games, which were drawn before endgame (material check is done, the number of played moves does not matter) and draws after the engine had a material advantage of at least 1 pawn during a game, because the engine should win a game, if material was won. All these bad draws are finally checked for a material disadvantage of at least 1 pawn: Because draws with material disadvantage prevented a possible loss and so, these games are no bad draws and are not counted.)
So, what we learn from my EAS-Tool about CSTal is: Number of played sacs are quite good in CSTal (EAS), but the engine plays not "straight forward" for fast wins and has problems avoiding early draws (draws before endgame)...
Uri Blass
Posts : 207 Join date : 2020-11-28
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 2:21 pm
pohl4711 wrote:
Chris Whittington wrote:
I'm quite disappointed with this version and draw rates. You can expect an upgrade which deals with the opening book plus the cores/threads issues, and with a more aggressive NNUE. Couple of weeks maybe.
As I mentioned before, playing aggressive is not only about playing sacs. This is why my EAS-Tool looks for short wins and bad draws and the average length of all won games of an engine (and adds points for all these things). And here we see, what is still not good in CSTal (EAS/normal):
For CSTal EAS: short wins percent-value is 11.2%, which is a really bad value for an engine, claiming an aggressive playing style. Also the average length of all won games is way too high: 82 moves... (76 moves is the overall average value in my SPCC-Ratinglist and other engines, which play aggressive, have values below 76 (Danasah 9: 70 moves, Komodo 14 aggressive 68 moves (!!!))). And the percentage of bad draws for CSTal EAS is 22.6%, which is too high. Most aggressive engines have bad draw percentages of 17% or below... (what are bad draws for my EAS-Tool? : Bad draws are games, which were drawn before endgame (material check is done, the number of played moves does not matter) and draws after the engine had a material advantage of at least 1 pawn during a game, because the engine should win a game, if material was won. All these bad draws are finally checked for a material disadvantage of at least 1 pawn: Because draws with material disadvantage prevented a possible loss and so, these games are no bad draws and are not counted.)
So, what we learn from my EAS-Tool about CSTal is: Number of played sacs are quite good in CSTal (EAS), but the engine plays not "straight forward" for fast wins and has problems avoiding early draws (draws before endgame)...
The assumption that you need to win with a material advantage of at least one pawn has no basis. There are positions when one pawn advantage is not enough to win against correct play but it still gives better practical chances than equal material.
pohl4711
Posts : 159 Join date : 2022-03-01 Location : Berlin
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 2:55 pm
Uri Blass wrote:
The assumption that you need to win with a material advantage of at least one pawn has no basis. There are positions when one pawn advantage is not enough to win against correct play but it still gives better practical chances than equal material.
For a statistic-tool like my EAS-Tool, the question is, what is true for most of the games/situations/positions. Not, what is true for some games/situations/positions.
mwyoung
Posts : 880 Join date : 2020-11-25 Location : USA
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 2:57 pm
pohl4711 wrote:
Chris Whittington wrote:
I'm quite disappointed with this version and draw rates. You can expect an upgrade which deals with the opening book plus the cores/threads issues, and with a more aggressive NNUE. Couple of weeks maybe.
As I mentioned before, playing aggressive is not only about playing sacs. This is why my EAS-Tool looks for short wins and bad draws and the average length of all won games of an engine (and adds points for all these things). And here we see, what is still not good in CSTal (EAS/normal):
For CSTal EAS: short wins percent-value is 11.2%, which is a really bad value for an engine, claiming an aggressive playing style. Also the average length of all won games is way too high: 82 moves... (76 moves is the overall average value in my SPCC-Ratinglist and other engines, which play aggressive, have values below 76 (Danasah 9: 70 moves, Komodo 14 aggressive 68 moves (!!!))). And the percentage of bad draws for CSTal EAS is 22.6%, which is too high. Most aggressive engines have bad draw percentages of 17% or below... (what are bad draws for my EAS-Tool? : Bad draws are games, which were drawn before endgame (material check is done, the number of played moves does not matter) and draws after the engine had a material advantage of at least 1 pawn during a game, because the engine should win a game, if material was won. All these bad draws are finally checked for a material disadvantage of at least 1 pawn: Because draws with material disadvantage prevented a possible loss and so, these games are no bad draws and are not counted.)
So, what we learn from my EAS-Tool about CSTal is: Number of played sacs are quite good in CSTal (EAS), but the engine plays not "straight forward" for fast wins and has problems avoiding early draws (draws before endgame)...
Is this post a joke.
Your assumptions are wrong.
"Bad draws are games, which were drawn before endgame (material check is done, the number of played moves does not matter) and draws after the engine had a material advantage of at least 1 pawn during a game, because the engine should win a game, if material was won."
And exactly how would you know this to be true........
Peter Berger
Posts : 131 Join date : 2020-11-20
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 3:23 pm
I really like one part of your decision for bad draws - it is about boredom.
I played exactly one game with CSTal thus far - in my usual setup against Crafty at tournament time control. CSTal equalized in a Berlin Spanish endgame with black, pretty easily. At about move 20 everything was traded, bishops of opposite colours - as I operate these games manually and it is a Sunday, I just stopped the game (that CSTal may still have won somehow eventually).
mwyoung
Posts : 880 Join date : 2020-11-25 Location : USA
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 3:34 pm
Peter Berger wrote:
I really like one part of your decision for bad draws - it is about boredom.
I played exactly one game with CSTal thus far - in my usual setup against Crafty at tournament time control. CSTal equalized in a Berlin Spanish endgame with black, pretty easily. At about move 20 everything was traded, bishops of opposite colours - as I operate these games manually and it is a Sunday, I just stopped the game (that CSTal may still have won somehow eventually).
I am not sure some people actually play the game of chess here. And understand the game of chess.
Chess is a two player game, and one player can not dictate the games style, game length, or draw rate.
These factors are only dictated by the quality of play of the other player.
And chess is not won by material advantages. Material advantage is just the usually outcome of a won position.
And losing material is also a way to draw the game of chess, and to win a game of chess.
Chess is the game of exceptions....
mwyoung
Posts : 880 Join date : 2020-11-25 Location : USA
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 5:46 pm
texium wrote:
Fascinating game where CSTal crushes leela in a sicilian with the black pieces when white leaves book winning. Sacrifices expected. First move out of book is move 14 [pgn] [Event "CSTal vs Leela T1"] [Site "Lucas Chess R 2.08a"] [Date "2023.07.01"] [White "Lc0 v0.30.0"] [Black "CSTal-NN-Strong"] [Result "0-1"] [WhiteElo "3700"] [BlackElo "3700"] [ECO "B78"] [TimeControl "180+2"] [PlyCount "144"] [Opening "Sicilian Defence: Dragon Variation, Yugoslav Attack, Old Line"] [Termination "Mate"]
The game is fascinating but not for the reasons you are stating.
I took a few hours to study this game to understand what happened.
And on move 14 maybe out of your book, but the position is know theory. And White is not winning in this positions. The computers did not leave know theory until move 24 with the move 24.Qg5??.
And it is highly unlikely Lc0 played this game of chess without some kind of major problem. Lc0 played weaker then many FM human players in this game. And played some outright blunders. And I could not replicate the major errors of Lc0 0.30.0 even at 1 second a move.
In this game Lc0 played with a error rate of .49 centipawns per move average and this is not even GM strength. No modern computer would play this badly. Where Chess System Tal had a flawless error rate of .02 centipawns per move average.
Posts : 2608 Join date : 2020-11-17 Location : Netherlands
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 6:37 pm
pohl4711 wrote:
Chris Whittington wrote:
I'm quite disappointed with this version and draw rates. You can expect an upgrade which deals with the opening book plus the cores/threads issues, and with a more aggressive NNUE. Couple of weeks maybe.
As I mentioned before, playing aggressive is not only about playing sacs. This is why my EAS-Tool looks for short wins and bad draws and the average length of all won games of an engine (and adds points for all these things). And here we see, what is still not good in CSTal (EAS/normal):
For CSTal EAS: short wins percent-value is 11.2%, which is a really bad value for an engine, claiming an aggressive playing style. Also the average length of all won games is way too high: 82 moves... (76 moves is the overall average value in my SPCC-Ratinglist and other engines, which play aggressive, have values below 76 (Danasah 9: 70 moves, Komodo 14 aggressive 68 moves (!!!))). And the percentage of bad draws for CSTal EAS is 22.6%, which is too high. Most aggressive engines have bad draw percentages of 17% or below... (what are bad draws for my EAS-Tool? : Bad draws are games, which were drawn before endgame (material check is done, the number of played moves does not matter) and draws after the engine had a material advantage of at least 1 pawn during a game, because the engine should win a game, if material was won. All these bad draws are finally checked for a material disadvantage of at least 1 pawn: Because draws with material disadvantage prevented a possible loss and so, these games are no bad draws and are not counted.)
So, what we learn from my EAS-Tool about CSTal is: Number of played sacs are quite good in CSTal (EAS), but the engine plays not "straight forward" for fast wins and has problems avoiding early draws (draws before endgame)...
Is there a (perhaps hidden) link on your website to download the pgn?
Dio
Posts : 222 Join date : 2021-08-28
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 6:45 pm
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 8:52 pm
mwyoung wrote:
texium wrote:
Fascinating game where CSTal crushes leela in a sicilian with the black pieces when white leaves book winning. Sacrifices expected. First move out of book is move 14 [pgn] [Event "CSTal vs Leela T1"] [Site "Lucas Chess R 2.08a"] [Date "2023.07.01"] [White "Lc0 v0.30.0"] [Black "CSTal-NN-Strong"] [Result "0-1"] [WhiteElo "3700"] [BlackElo "3700"] [ECO "B78"] [TimeControl "180+2"] [PlyCount "144"] [Opening "Sicilian Defence: Dragon Variation, Yugoslav Attack, Old Line"] [Termination "Mate"]
The game is fascinating but not for the reasons you are stating.
I took a few hours to study this game to understand what happened.
And on move 14 maybe out of your book, but the position is know theory. And White is not winning in this positions. The computers did not leave know theory until move 24 with the move 24.Qg5??.
And it is highly unlikely Lc0 played this game of chess without some kind of major problem. Lc0 played weaker then many FM human players in this game. And played some outright blunders. And I could not replicate the major errors of Lc0 0.30.0 even at 1 second a move.
In this game Lc0 played with a error rate of .49 centipawns per move average and this is not even GM strength. No modern computer would play this badly. Where Chess System Tal had a flawless error rate of .02 centipawns per move average.
I'm sorry but you're trying to use another engine ( presumably SF15.1 to analyze Leela which is a stronger engine. I used Leela 0.30 with the net T1, the strongest net 768x15 which is stronger than BT2 and BT3, and which data will soon be used, if it already isn't, to further make stockfish stronger. I'll reply with a screenshot of the lines Leela plays in that position. They left the provided opening book at move 14 though.
mwyoung
Posts : 880 Join date : 2020-11-25 Location : USA
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 8:57 pm
texium wrote:
mwyoung wrote:
texium wrote:
Fascinating game where CSTal crushes leela in a sicilian with the black pieces when white leaves book winning. Sacrifices expected. First move out of book is move 14 [pgn] [Event "CSTal vs Leela T1"] [Site "Lucas Chess R 2.08a"] [Date "2023.07.01"] [White "Lc0 v0.30.0"] [Black "CSTal-NN-Strong"] [Result "0-1"] [WhiteElo "3700"] [BlackElo "3700"] [ECO "B78"] [TimeControl "180+2"] [PlyCount "144"] [Opening "Sicilian Defence: Dragon Variation, Yugoslav Attack, Old Line"] [Termination "Mate"]
The game is fascinating but not for the reasons you are stating.
I took a few hours to study this game to understand what happened.
And on move 14 maybe out of your book, but the position is know theory. And White is not winning in this positions. The computers did not leave know theory until move 24 with the move 24.Qg5??.
And it is highly unlikely Lc0 played this game of chess without some kind of major problem. Lc0 played weaker then many FM human players in this game. And played some outright blunders. And I could not replicate the major errors of Lc0 0.30.0 even at 1 second a move.
In this game Lc0 played with a error rate of .49 centipawns per move average and this is not even GM strength. No modern computer would play this badly. Where Chess System Tal had a flawless error rate of .02 centipawns per move average.
I'm sorry but you're trying to use another engine ( presumably SF15.1 to analyze Leela which is a stronger engine. I used Leela 0.30 with the net T1, the strongest net 768x15 which is stronger than BT2 and BT3, and which data will soon be used, if it already isn't, to further make stockfish stronger. I'll reply with a screenshot of the lines Leela plays in that position. They left the provided opening book at move 14 though.
Error rate is the Error rate. And no chess engine today plays at a FM human level.
And why the losing move 24.Qg5. Leela does not want to play that move.
Mclane
Posts : 3022 Join date : 2020-11-17 Age : 57 Location : United States of Europe, Germany, Ruhr area
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 8:57 pm
results of my 8 core machine, permanent brain worked;
924 Partien gespielt / Turnier beendet Name des Turniers: Ryzen7 5700 Ort/ Land: LAPTOP-JTQQP8M7, Deutschland Spielstufe: Turnier 40/5 Hardware: AMD Ryzen 7 5700U with Radeon Graphics mit 4.096 MB Speicher
TheSelfImprover likes this post
texium
Posts : 125 Join date : 2022-07-19
Subject: Re: Chess System Tal 2 released Sun Jul 02, 2023 8:58 pm
in this position, the current leela net at almost 100k nodes over 1min, evaluated Nb3 as the best move, it doesn't realize until it plays it, that Qa3+ is a valid option and then says that instead of white completely win, black has a slight edge. As this was played in a blitz game, the engine wouldn't be thinking for 20 sec and so leela evaluated Qg5 as the best move here: . I had them playing a 3+2 game.
Home 
