Chess System Tal 2.1 EAS

[Rebel]

Chris asked me to announce the new Chess System Tal 2.1 EAS

The new EAS version of Chess System Tal has greatly improved when it is about playing style playing unbelievable sharp games with lots of material sacrifices, Paul Morphy would have blushed. Stefan Pohl kindly tested Chess System Tal 2.1 EAS already, here are the results.

Code: Select all

EAS (Top10 of my full EAS-Ratinglist):
                                 bad  avg.win 
Rank  EAS-Score  sacs   shorts  draws  moves  Engine/player 
-------------------------------------------------------------------
   1    429481  51.61%  38.03%  05.49%   66   Patricia 3.1 avx2  
   2    376711  50.62%  36.92%  07.20%   68   Chess System Tal 2.1 EAS  
   3    337432  46.96%  29.39%  03.46%   70   Patricia 4.0 avx2  
   4    312346  43.56%  27.63%  04.21%   68   Cerberus 21124081r81  
   5    250682  37.88%  23.88%  08.79%   75   Rebel Extreme avx2  
   6    218083  32.46%  23.04%  09.25%   71   Velvet 8.1.0 risky  
   7    201351  34.75%  22.22%  10.27%   73   Velvet 8.0.0 risky  
   8    197919  31.18%  29.46%  17.09%   71   Revenge 1.0 avx2  
   9    193748  21.29%  26.40%  09.17%   70   Stockfish 16 230630  
  10    190732  22.08%  24.73%  08.71%   70   Stockfish 16.1 240224

Link

[Graham Banks]

Thanks, Chris.

[fkarger]

Cool

[pohl4711]

I tested this engine as an experimental dev of Rebel (Rebel Extreme E145). Because it was an experimental testrun, I did not include the results in my full ratinglist and EAS-ratinglist permanently.

Will include the games right now as games by CSTal 2.1 EAS in my ratinglist.

EAS-calculation takes 4.5 hours for my full ratinglist, so please allow 1-2 days of delay.

The bottom of the full ratinglist (Celo) will look like this

Code: Select all

      Program                     Elo     +    - Games    Score   Av.Op. Draws
 109 Stockfish final HCE       : 3468    3    3 23000    43.7%   3516   38.4%
 110 Slow Chess 2.9 avx2       : 3448    4    4 23000    41.8%   3511   42.5%
 111 Cerberus 21124081r81      : 3447    4    4 14000    39.9%   3520   37.4%
 112 Rebel Extreme avx2        : 3402    4    4 14000    34.0%   3520   39.3%
 113 Patricia 4.0 avx2         : 3381    4    4 14000    31.4%   3520   32.2%
 114 Revenge 1.0 avx2          : 3356    5    5 15000    18.3%   3628   30.3%
 115 CSTal 2.1 EAS             : 3343    4    4 14000    26.9%   3520   29.6%
 116 Monty 241220 a512         : 3321    5    5 14000    24.6%   3520   35.4%
 117 Patricia 3.1 avx2         : 3209    6    6 14000    14.7%   3520   21.1%
 

[chrisw]

I tested this engine as an experimental dev of Rebel (Rebel Extreme E145). Because it was an experimental testrun, I did not include the results in my full ratinglist and EAS-ratinglist permanently.

Will include the games right now as games by CSTal 2.1 EAS in my ratinglist.

EAS-calculation takes 4.5 hours for my full ratinglist, so please allow 1-2 days of delay.

Would be fun to have a panel (reasonably strong humans?) which took a few selected games from the strong EAS programs and awarded (somehow) a brilliancy prize or two.

A partial automated approach would/could be to select some games by two joint criteria, first the amount of material sacrificed (as you do already) and the length of moves where the successful sacrificer remains in material deficit for converting the win (do you do this already? Presumably)

One thing I would guess you don’t do is select positions where material is nominally hanging but can’t be (or isn’t) taken. Many Tal games were like this, stuff continually hanging all over the place. Probably difficult with your software though?

[pohl4711]

A partial automated approach would/could be to select some games by two joint criteria, first the amount of material sacrificed (as you do already) and the length of moves where the successful sacrificer remains in material deficit for converting the win (do you do this already? Presumably)

One thing I would guess you don’t do is select positions where material is nominally hanging but can’t be (or isn’t) taken. Many Tal games were like this, stuff continually hanging all over the place. Probably difficult with your software though?

I can not do this.
I use pgn-extract and just look for won games, where the winning color had less material for (at least) 8 consecutive plies. If this is true, it is a sac-game for my EAS-tool.
Very simple. But very fast and it works (most of the time) in enginegames, because engines do not blunder: So, if an engine has less material in a game, but wins the game at the end, the engine has played a sac.

[pohl4711]

Would be fun to have a panel (reasonably strong humans?) which took a few selected games from the strong EAS programs and awarded (somehow) a brilliancy prize or two.

For this, I recommend using my Interesting Wins Search Tool. It uses the same sac-search as the EAS-tool, but does no statistics, so it is way faster. And it sorts the games better.
https://www.sp-cc.de/files/interesting_ ... ch_tool.7z

Here some details from the ReadMe:
The 2 output-files of the IWS-Tool contain different types of wins:

interesting_wins.pgn contains:
1) Queen Sacrifices, followed by
2) 5+ PawnUnit Sacrifices, followed by
3) 4 PawnUnit Sacrifices, followed by
4) 3 PawnUnit Sacrifices, followed by
5) 2 PawnUnit Sacrifices, followed by
6) 1 PawnUnit Sacrifices, followed by
7) Games, ended before endgame (material) was reached, followed by
8) Games with material imbalance (Rook vs. Bishop and 2 pawns for example)

very_interesting_wins.pgn contains: same as above, but without category 6 and 8 (1 PawnUnit sacs and Material
imbalance), because, if the IWS-Tool is used on huge databases, the interesting_wins.pgn-file can be very
huge, too.

The games in the output-files are sorted in 2 ways:
First: The games are sorted by categories (category 1 is followed by category 2, 3, ... etc.).
Second: In each category, the games are sorted by length (0-19 moves, followed by 20-29 moves, followed by
30-39 moves... and so on, up to 120 moves and beyond). So, in each category, the shortest wins are at the
beginning and followed by the longer wins...

And, there are no double games in one output-file: If a game fits more than one category, it is stored
in the lowest category, all other apperances of this game in higher categories are deleted.
For example: A game contains a 3 PawnUnit-Sacrifice and is won before the endgame material is reached:
This game is stored in category 4 (= 3 PawnUnit Sacrifices) and not in category 7...

And each games gets a new Annotator-Tag, so it is clear, which category the game belongs to.
One of these 8 tags is added to each game-notation:

[Annotator "IWS-Tool: Queen Sacrifice found in this game"]
[Annotator "IWS-Tool: 5+ PawnUnits Sacrifice found in this game"]
[Annotator "IWS-Tool: 4 PawnUnits Sacrifice found in this game"]
[Annotator "IWS-Tool: 3 PawnUnits Sacrifice found in this game"]
[Annotator "IWS-Tool: 2 PawnUnits Sacrifice found in this game"]
[Annotator "IWS-Tool: 1 PawnUnit Sacrifice found in this game"]
[Annotator "IWS-Tool: Game ended before endgame (material)"]
[Annotator "IWS-Tool: Material imbalance found in this game"]

[reflectionofpower]

Thx Chris

[cpeters]

Hello!

Is this system windows-only?

greetings

[Eelco de Groot]

Hello!

Is this system windows-only?

greetings

I think so yes, all CS Tals and Rebels were built with MSVC in the past as that is Chris' compiler of choice and other compilers had trouble with the very large networks used at least that's what I remember. The .exe is uncompressed 139 Mb and Google could not viruscheck something that large :0 (so you have to give Google permission to download anyway). Malwarebytes scanned it okay and runs okay in Shredder GUI here:

Code: Select all

Chess System Tal 2.1 EAS
Release version
(C) Chris Whittington 2003-2025
NNUE development Ed Schroder 2025
==================================
GenuineIntel
Intel(R) Core(TM) i5-4440 CPU @ 3.10GHz
3DNOW not supported
3DNOWEXT not supported
ABM not supported
ADX not supported
AES supported
AVX supported
AVX2 supported
AVX512CD not supported
AVX512ER not supported
AVX512F not supported
AVX512PF not supported
BMI1 supported
BMI2 supported
CMPXCHG16B supported
CX8 supported
ERMS supported
F16C supported
FMA supported
FSGSBASE supported
FXSR supported
HLE not supported
INVPCID supported
LAHF supported
LZCNT supported
MMX supported
MMXEXT not supported
MONITOR supported
MOVBE supported
MSR supported
OSXSAVE supported
PCLMULQDQ supported
POPCNT supported
PREFETCHWT1 not supported
RDRAND supported
RDSEED not supported
RDTSCP supported
RTM not supported
SEP supported
SHA not supported
SSE supported
SSE2 supported
SSE3 supported
SSE4.1 supported
SSE4.2 supported
SSE4a not supported
SSSE3 supported
SYSCALL supported
TBM not supported
XOP not supported
XSAVE supported
Processor supports AVX2
Hash=128 Mb
Globals=199 Kb
Threads=1
Cores=2
Hamming=14
Hamming=15
Thread=74 Mb
Tree=7 Mb
Magics=862720 bytes
set Hash to 1024MB
Hash table set to 1024 MB
set Ponder to value true
Engine: Chess System Tal 2.1 EAS (1024 MB)
gemaakt door Chris Whittington (C) 2003-2025