CSTal-Elo, CSTal-EAS, CSTal-EXTREME and CSTal-ABSURD. Released

[abgursu]

Hello sir, thanks for these releases.
Something I don't understand is, both the 2.07 EAS version and 2.00 EAS version share the same net: 1162. I wonder what is the primary difference in their size tho, 67 MB for 2.07 versions but 135 MB for 2.00 Versions - with the same net - which was almost double the size. Can you share a word to explain please?

[chrisw]

It’s because it was using a very expensive and dumb way to compile the net inline. Now is more sensible. 50% sounds about right.

[pohl4711]

V2.07 Absurd avx2 seem to be totally broken. I started a testmatch vs. 2.06 Absurd avx2 (singlethread, 512 Hash, Balanced HERT openings, 60s+0.5s thinking-time). Hardware: Ryzen 9 3900 12 core CPU, only 12 games simultaneously, no "overload" (which I know, that it is no problem (but you dont), so I wanted to avoid it here):

After 50 games, 2.07 lost 42 of 50 games (7 draws, 1 win) vs 2.06 !!! So, I aborted the testrun.

Here one of the games, if you look at the search depths, the 2.07 has around -2 plies less than 2.06 (this would explain the bad result).

[Date "2026.07.04"]
[Round "1"]
[White "CS Tal Absurd 2.06"]
[Black "CS Tal Absurd 2.07"]
[Result "1-0"]
[ECO "B42"]
[GameDuration "00:02:30"]
[GameEndTime "2026-07-04T07:45:05.997 Mitteleuropäische Sommerzeit"]
[GameStartTime "2026-07-04T07:42:35.207 Mitteleuropäische Sommerzeit"]
[Opening "Sicilian"]
[PlyCount "105"]
[TimeControl "60+0.5"]
[Variation "Kan, 5.Bd3"]

1. e4 {book} c5 {book} 2. Nf3 {book} e6 {book} 3. d4 {book} cxd4 {book}
4. Nxd4 {book} a6 {book} 5. Bd3 {book} Nf6 {book} 6. O-O {+0.46/19 3.5s}
d6 {-0.38/18 2.5s} 7. c4 {+0.51/20 1.6s} Be7 {-0.46/17 2.1s}
8. Nc3 {+0.55/20 2.8s} O-O {-0.42/20 4.9s} 9. Be3 {+0.60/20 2.2s}
Nbd7 {-0.39/19 4.2s} 10. f4 {+0.59/18 2.9s} Re8 {-0.44/17 4.0s}
11. Bc2 {+0.51/18 5.7s} Bf8 {-0.38/16 1.7s} 12. h3 {+0.56/19 2.5s}
b6 {-0.39/16 4.2s} 13. e5 {+0.59/18 3.2s} Bb7 {-1.02/15 2.8s}
14. exf6 {+2.11/18 1.2s} Nxf6 {-1.02/18 9.7s} 15. Ba4 {+2.32/19 2.5s}
Rc8 {-0.93/14 2.0s} 16. Bxe8 {+2.64/16 2.0s} Qxe8 {-1.02/15 2.6s}
17. Qe2 {+2.69/17 1.7s} h6 {-0.99/15 2.8s} 18. Rac1 {+2.68/16 1.3s}
Qd7 {-1.01/15 7.1s} 19. Rfd1 {+2.79/16 2.2s} Re8 {-1.10/13 1.8s}
20. Nf3 {+2.92/17 3.8s} Qd8 {-1.13/12 0.75s} 21. Qf2 {+2.97/17 1.5s}
Qb8 {-1.26/14 2.3s} 22. Nd2 {+2.89/17 1.8s} a5 {-1.15/13 1.3s}
23. Bxb6 {+3.26/16 1.4s} Be7 {-1.38/13 1.5s} 24. Nb5 {+3.36/17 1.5s}
Rc8 {-1.12/11 0.82s} 25. Ba7 {+3.35/17 0.95s} Qa8 {-1.27/13 0.83s}
26. Kh2 {+3.36/16 1.6s} a4 {-1.14/13 1.0s} 27. b3 {+3.40/16 1.0s}
Kh7 {-1.26/13 2.2s} 28. Bd4 {+3.39/14 1.5s} Rg8 {-1.26/13 2.0s}
29. Rg1 {+3.50/16 2.0s} axb3 {-1.44/11 0.68s} 30. axb3 {+3.37/16 0.89s}
Rd8 {-1.43/11 0.62s} 31. Ra1 {+3.75/18 1.8s} Qc8 {-1.62/14 1.1s}
32. Bb6 {+3.73/17 0.76s} Re8 {-1.41/12 0.73s} 33. Bc7 {+4.08/15 1.1s}
e5 {-1.31/13 1.2s} 34. Bxd6 {+4.39/14 0.73s} e4 {-1.59/10 0.63s}
35. Rge1 {+4.46/16 1.3s} Bd8 {-1.33/10 0.43s} 36. c5 {+4.39/14 1.4s}
Qd7 {-1.96/12 0.70s} 37. Nd4 {+5.24/14 0.95s} Ng4+ {-2.49/11 0.63s}
38. hxg4 {+5.72/14 0.83s} Bf6 {-2.60/14 1.6s} 39. Nf5 {+6.20/15 1.3s}
e3 {-3.13/11 0.47s} 40. Nxe3 {+6.20/14 0.29s} Bc3 {-3.40/11 0.81s}
41. Ra7 {+7.29/15 1.3s} Qb5 {-4.06/11 1.1s} 42. Nf5 {+9.03/15 1.5s}
Ba8 {-5.17/11 0.59s} 43. Rxf7 {+10.85/16 1.1s} Bd5 {-6.12/10 0.24s}
44. Rxe8 {+12.10/14 0.24s} Qxe8 {-8.93/13 0.85s} 45. Re7 {+12.73/15 0.73s}
Qxe7 {-9.01/10 0.090s} 46. Nxe7 {+15.08/13 0.68s} Bc6 {-11.01/11 0.69s}
47. g5 {+M25/17 0.80s} g6 {-13.27/11 0.38s} 48. f5 {+M13/32 0.36s}
Kg7 {-M20/13 0.40s} 49. Qe3 {+M9/43 0.41s} hxg5 {-M10/24 0.12s}
50. Qxc3+ {+M7/51 0.29s} Kf7 {-M6/44 0.085s} 51. Qh8 {+M5/82 0.18s}
gxf5 {-M4/97 0.041s} 52. Qg8+ {+M3/116 0.017s} Kf6 {-M2/116 0.013s}
53. Qg6# {+M1/116 0.004s, White mates} 1-0

[chrisw]

V2.07 Absurd avx2 seem to be totally broken. I started a testmatch vs. 2.06 Absurd avx2 (singlethread, 512 Hash, Balanced HERT openings, 60s+0.5s thinking-time). Hardware: Ryzen 9 3900 12 core CPU, only 12 games simultaneously, no "overload" (which I know, that it is no problem (but you dont), so I wanted to avoid it here):

After 50 games, 2.07 lost 42 of 50 games (7 draws, 1 win) vs 2.06 !!! So, I aborted the testrun.

Here one of the games, if you look at the search depths, the 2.07 has around -2 plies less than 2.06 (this would explain the bad result).

[Date "2026.07.04"]
[Round "1"]
[White "CS Tal Absurd 2.06"]
[Black "CS Tal Absurd 2.07"]
[Result "1-0"]
[ECO "B42"]
[GameDuration "00:02:30"]
[GameEndTime "2026-07-04T07:45:05.997 Mitteleuropäische Sommerzeit"]
[GameStartTime "2026-07-04T07:42:35.207 Mitteleuropäische Sommerzeit"]
[Opening "Sicilian"]
[PlyCount "105"]
[TimeControl "60+0.5"]
[Variation "Kan, 5.Bd3"]

1. e4 {book} c5 {book} 2. Nf3 {book} e6 {book} 3. d4 {book} cxd4 {book}
4. Nxd4 {book} a6 {book} 5. Bd3 {book} Nf6 {book} 6. O-O {+0.46/19 3.5s}
d6 {-0.38/18 2.5s} 7. c4 {+0.51/20 1.6s} Be7 {-0.46/17 2.1s}
8. Nc3 {+0.55/20 2.8s} O-O {-0.42/20 4.9s} 9. Be3 {+0.60/20 2.2s}
Nbd7 {-0.39/19 4.2s} 10. f4 {+0.59/18 2.9s} Re8 {-0.44/17 4.0s}
11. Bc2 {+0.51/18 5.7s} Bf8 {-0.38/16 1.7s} 12. h3 {+0.56/19 2.5s}
b6 {-0.39/16 4.2s} 13. e5 {+0.59/18 3.2s} Bb7 {-1.02/15 2.8s}
14. exf6 {+2.11/18 1.2s} Nxf6 {-1.02/18 9.7s} 15. Ba4 {+2.32/19 2.5s}
Rc8 {-0.93/14 2.0s} 16. Bxe8 {+2.64/16 2.0s} Qxe8 {-1.02/15 2.6s}
17. Qe2 {+2.69/17 1.7s} h6 {-0.99/15 2.8s} 18. Rac1 {+2.68/16 1.3s}
Qd7 {-1.01/15 7.1s} 19. Rfd1 {+2.79/16 2.2s} Re8 {-1.10/13 1.8s}
20. Nf3 {+2.92/17 3.8s} Qd8 {-1.13/12 0.75s} 21. Qf2 {+2.97/17 1.5s}
Qb8 {-1.26/14 2.3s} 22. Nd2 {+2.89/17 1.8s} a5 {-1.15/13 1.3s}
23. Bxb6 {+3.26/16 1.4s} Be7 {-1.38/13 1.5s} 24. Nb5 {+3.36/17 1.5s}
Rc8 {-1.12/11 0.82s} 25. Ba7 {+3.35/17 0.95s} Qa8 {-1.27/13 0.83s}
26. Kh2 {+3.36/16 1.6s} a4 {-1.14/13 1.0s} 27. b3 {+3.40/16 1.0s}
Kh7 {-1.26/13 2.2s} 28. Bd4 {+3.39/14 1.5s} Rg8 {-1.26/13 2.0s}
29. Rg1 {+3.50/16 2.0s} axb3 {-1.44/11 0.68s} 30. axb3 {+3.37/16 0.89s}
Rd8 {-1.43/11 0.62s} 31. Ra1 {+3.75/18 1.8s} Qc8 {-1.62/14 1.1s}
32. Bb6 {+3.73/17 0.76s} Re8 {-1.41/12 0.73s} 33. Bc7 {+4.08/15 1.1s}
e5 {-1.31/13 1.2s} 34. Bxd6 {+4.39/14 0.73s} e4 {-1.59/10 0.63s}
35. Rge1 {+4.46/16 1.3s} Bd8 {-1.33/10 0.43s} 36. c5 {+4.39/14 1.4s}
Qd7 {-1.96/12 0.70s} 37. Nd4 {+5.24/14 0.95s} Ng4+ {-2.49/11 0.63s}
38. hxg4 {+5.72/14 0.83s} Bf6 {-2.60/14 1.6s} 39. Nf5 {+6.20/15 1.3s}
e3 {-3.13/11 0.47s} 40. Nxe3 {+6.20/14 0.29s} Bc3 {-3.40/11 0.81s}
41. Ra7 {+7.29/15 1.3s} Qb5 {-4.06/11 1.1s} 42. Nf5 {+9.03/15 1.5s}
Ba8 {-5.17/11 0.59s} 43. Rxf7 {+10.85/16 1.1s} Bd5 {-6.12/10 0.24s}
44. Rxe8 {+12.10/14 0.24s} Qxe8 {-8.93/13 0.85s} 45. Re7 {+12.73/15 0.73s}
Qxe7 {-9.01/10 0.090s} 46. Nxe7 {+15.08/13 0.68s} Bc6 {-11.01/11 0.69s}
47. g5 {+M25/17 0.80s} g6 {-13.27/11 0.38s} 48. f5 {+M13/32 0.36s}
Kg7 {-M20/13 0.40s} 49. Qe3 {+M9/43 0.41s} hxg5 {-M10/24 0.12s}
50. Qxc3+ {+M7/51 0.29s} Kf7 {-M6/44 0.085s} 51. Qh8 {+M5/82 0.18s}
gxf5 {-M4/97 0.041s} 52. Qg8+ {+M3/116 0.017s} Kf6 {-M2/116 0.013s}
53. Qg6# {+M1/116 0.004s, White mates} 1-0

Why don't you research first, or is your idea to trash everything every time, whoops you didn't mention your pet bias project here, overwhelming your ability to work it out. It's already been reported. CSTal206 absurd shipped with the wrong net, which wasn't absurd at all, CSTal207 shipped with the absurd net and fixed it. Shock horror!! Pohl discovers that a strong (wrong) net in CSTal206 (not absurd at all) defeats an absurd net in CSTal207 and calls it "broken". Strong net beats weak net. Doh!! Why are you even testing overwritten and hence withdrawn releases? Or you just want to make up some nonsense to complain about? Shall I complain about your appallingly badly written, unmaintainable, broken for several years EAS tester or the problem with testers getting all biased because they favour particular engines and dislike others? Or maybe you dislike the programmers as your disgusting posts on the German chess forum suggest? Should we wonder why Stockfish scores lower EAS than your Pet engine in your self-written, broken test when Stockfish has way more sacrifices? Could it be because you increased to literally insane levels the weighting of one parameter that totally outweighs all sacrifices but just so happens to hurl Pet engine to the top of your list? Your bias and weird creative self-selected weightings destroy your credibility as a chess engine tester. Either test and stop with the hate, bias and "creative" tests or continue testing, STFU on the bias and hate and quit promoting one engine and making fancy fake tests to "help" it. Testers are expected to be kind of neutral else they lose all credibility. As is clear from your foul posts on German chess forum, neutral you are certainly not.

[pohl4711]

Why are you releasing engines using the wrong net without testing them before? It is my fault, that I missed the bug report? Yes. Did I do the buggy engine release? No. You did.

Now all here can see it by themselves: You are just a small man with bad manners, consumed by envy that he didn't come up with the EAS tool himself, which made all the great engines, playing aggressive and spectacular (Patricia, Rebel Extreme, CS Tal) possible.