Stockfish 16 released

[syzygy]

Releasing SF 16 on June 30, and then having two NNUE improvements and a code improvement just one day later suggests some disorganization and poor communication within the SF author group. Surely with these big advancements in the pipeline it would have made more sense to wait one day, finalize the tests of the improvements, and only then introduce the latest product as SF 16.

Still, I don't mean to be unduly critical of the SF authors as they do a fine job in crafting this chess engine. However, some disorganization seems evident. So, perhaps they'd benefit from a Team Leader like they once had.

You are effectively arguing that SF development should be stopped for a month after the release of a new version.

[syzygy]

Yes. However, the difference doesn't matter much if you play with today's normal hardware with 6 cores and more. I've already played a few hundred games blitz on PlayChess with the new network, using my old 8-core Ryzen 2700. Runs great. I think the new size will continue to be trained and improved. I have no need to go back to the old size (SF 16).

Also, blitz on PlayChess has much longer time controls than STC on fishtest.

[Eduard]

I'm testing the big new 2048 network. Is it better than Stockfish 16? Positionally yes, tactically no, not only in the following position. I test some critical openings on PlayChess, Elos are unimportant to me. My opponent played with 35,000 kns and Stockfish 16, I played with 6000 kns and my Stockfish clone and the new 2048 large network.

See the difference analyzed here on Ryzen 2700:

TarasUA,Stockfish 16 - SolistaChess,Diamond 1 1-0 0.0, Wertungspartie, 5 min 2023
[fen]r2qk2r/ppp1n2p/5pp1/nPb5/2BN1B2/P1N1P3/5PPP/R2QK2R b KQkq - 0 1[/fen]


Analysis by Stockfish dev 010723-avx2:

14...Sxc4 15.0-0 Lxd4 16.exd4 Sb6 17.Lh6 Sbd5 18.Tc1 Kf7 19.Sxd5 Sxd5 20.Db3 Tc8 21.Tc5 c6 22.h4 Dd7 23.bxc6 bxc6 24.Tfc1 Thd8 25.Da4 Se7 26.Te1 Db7 27.Ta5 Td7 28.Dc4+ Sd5 29.Tc5 Td6 30.Tc1 De7 31.Lf4 Td7 32.Ld2 Td6 33.h5 De6 34.Ta5 Tc7 35.h6 Df5 36.Lb4 Tdd7 37.Te1 g5 38.De2 Td8 39.Tc5 Tdc8 40.La5 Tb7 41.Dh5+ Dg6 42.Dg4
Weiß steht besser: +/- (0.72 --) Tiefe: 43/82 00:04:14 1269MN, tb=11563
14...Sxc4 15.Db3
Weiß steht besser: +/- (0.78 --) Tiefe: 43/82 00:04:27 1330MN, tb=11854
14...Sxc4 15.Db3
Weiß steht besser: +/- (0.87 --) Tiefe: 43/82 00:04:29 1343MN, tb=11880
14...Sxc4 15.Db3
Weiß steht besser: +/- (0.98 --) Tiefe: 43/82 00:04:47 1426MN, tb=12928
14...Sxc4 15.Db3
Weiß steht besser: +/- (1.13 --) Tiefe: 43/82 00:04:54 1460MN, tb=13849

I rebooted the PC and then started the analysis with Stockfish 16:

TarasUA,Stockfish 16 - SolistaChess,Diamond 1 1-0 0.0, Wertungspartie, 5 min 2023
r2qk2r/ppp1n2p/5pp1/nPb5/2BN1B2/P1N1P3/5PPP/R2QK2R b KQkq - 0 1

Analysis by Stockfish 16-avx2:

14...Sxc4 15.0-0 Lxd4 16.exd4 Sb6 17.Lh6 Sbd5 18.Tc1 Kf7 19.Db3 Te8 20.Tfe1 c6 21.Se4 Sf5 22.bxc6 bxc6 23.Db7+ Dc7 24.Dxc7+ Sxc7 25.Txc6 Sd5 26.Ld2 Tad8 27.Kf1 g5 28.Ta6 Td7 29.Sc5 Txe1+ 30.Kxe1 Te7+ 31.Kd1 h5 32.a4 h4 33.Sb3 Tc7 34.Ta5 Sfe7 35.Ke2 Sc3+ 36.Kd3 Sd1
Weiß steht etwas besser: +/= (0.66 ++) Tiefe: 37/83 00:01:20 528MN, tb=1589
14...Sxc4 15.Db3
Weiß steht besser: +/- (0.84 --) Tiefe: 37/83 00:01:24 556MN, tb=1603
14...Sxc4 15.Db3
Weiß steht besser: +/- (0.95 --) Tiefe: 37/83 00:01:27 574MN, tb=1603
14...Sxc4 15.Db3
Weiß steht besser: +/- (1.10 --) Tiefe: 37/83 00:01:31 594MN, tb=1615
14...Sxc4 15.Db3
Weiß steht deutlich besser: +/- (1.31 --) Tiefe: 37/83 00:01:37 631MN, tb=2776
14...Sxc4 15.Db3 Lxd4 16.0-0-0 Sa5 17.Da4 Lxc3 18.Txd8+ Txd8 19.Kc2 Sd5 20.b6+ Sc6 21.bxa7 Sb6 22.Db3 Le5 23.De6+ Se7 24.Lxe5 fxe5 25.Dxe5 Tf8 26.Kb1 Td6 27.Tc1 Tf5 28.Dh8+ Kd7 29.a8T Sxa8 30.Dxa8 Tb5+ 31.Ka2 Td2+ 32.Ka1 Tbb2 33.Dh8 Ta2+ 34.Kb1 Txa3 35.Tc4 Tad3 36.Db8 Sd5 37.Df8 Se7 38.Df4 Td6 39.e4 Sc6 40.Tc2 Td1+ 41.Kb2 h5 42.Df7+ Se7

To find the best move for White, the 2048 network needs 4min 27s. Stockfish 16 takes 1min 24s.

Game:

[pgn][Event "Wertungspartie, 5 min"]
[Site "Maschinenraum"]
[Date "2023.07.03"]
[Round "?"]
[White "TarasUA, Stockfish 16"]
[Black "SolistaChess, Diamond 1"]
[Result "1-0"]
[ECO "D08"]
[WhiteElo "3105"]
[BlackElo "3121"]
[Annotator "0.77;0.67"]
[PlyCount "175"]
[EventDate "2023.07.03"]
[SourceTitle "playchess.com"]
[TimeControl "300"]

{Stockfish 16 (30 threads): 41.0 plies; 35.367kN/s AMD Ryzen 9 5950X 16-Core Processor 3394MHz, (16 cores, 32 threads), Diamond.ctg, 1024 MB} 1. d4 {B 0} d5 {B 0} 2. c4 {B 0} e5 {B 0} 3. dxe5 {B 0} d4 {B 0} 4. Nf3 {B 0} Nc6 {B 0} 5. a3 {B 0} g6 {B 0} 6. Bg5 {B 0} Qd7 {B 0} 7. Bf4 {B 0} f6 {B 0} 8. e6 {0.77/37 12} Qd8 {B 0} 9. b4 {0.81/32 6} Bxe6 {B 0 (a6)} 10. b5 {0.82/32 4} Na5 {B 0} 11. Nxd4 {0.74/33 8} Bxc4 {B 0} 12. Nc3 {0.81/35 16} Bc5 {0.67/31 28} 13. e3 {0.79/38 0} Ne7 {0.67/27 20} 14. Bxc4 {1.39/39 3} Nxc4 {0.64/26 1} 15. Qb3 {1.44/33 3 (0-0)} Bxd4 {0.20/26 4} 16. O-O-O {1.52/34 1 (Dxc4)} Na5 {1.26/25 11} 17. Qa4 {1.57/36 4} Bxc3 {1.23/28 1} 18. Rxd8+ {1.61/37 6} Rxd8 {1.16/28 0} 19. Kc2 {1.69/37 4} Nd5 {1.07/29 2} 20. b6+ {1.69/35 3} c6 {1.06/29 0} 21. bxa7 {1.70/37 11} O-O {1.35/28 3 (a8)} 22. Bb8 {1.64/34 3} f5 {1.54/31 27 (b6)} 23. Rd1 {1.81/39 8} b5 {1.61/30 0} 24. Rxd5 {2.07/36 3} Rxd5 {1.81/28 6} 25. Qf4 {2.00/40 2} Be1 {1.84/30 4} 26. e4 {2.20/37 0} Rd2+ {1.94/30 9} 27. Qxd2 {2.35/44 0} Bxd2 {1.93/33 7} 28. a8=Q {2.35/48 0} Bf4 {1.81/31 3} 29. Qxa5 {2.35/50 2} Bxb8 {1.60/30 1} 30. f3 {2.35/48 4} Bxh2 {1.73/31 5 (fxe4)} 31. Qa6 {2.49/38 4 (Db4)} fxe4 {1.79/35 32} 32. fxe4 {2.65/50 1} Rf7 {2.01/27 4} 33. Qc8+ {2.65/41 0} Kg7 {2.03/28 7} 34. Qe6 {2.63/48 1} h5 {2.23/27 8 (g3)} 35. Kb3 {2.82/34 4} c5 {2.30/27 0} 36. Qb6 {2.82/40 3} c4+ {2.43/26 3} 37. Kc2 {3.01/37 1} h4 {2.64/28 5} 38. Qd4+ {3.47/36 2} Kh6 {2.61/28 1} 39. e5 {3.48/40 2} Kg5 {2.80/30 3} 40. Qd8+ {3.76/39 5} Kf5 {2.78/29 0} 41. Qe8 {4.08/36 5} Rc7 {2.90/28 8} 42. e6 {4.11/31 0} Kf6 {3.14/28 8 (g3)} 43. Qxb5 {4.92/33 11} g5 {3.57/26 6 (e7)} 44. a4 {5.44/29 4} Re7 {4.22/24 12 (f4)} 45. a5 {6.16/31 2} Bg3 {4.27/24 4 (g1)} 46. Qb2+ {6.33/27 6 (Kd1)} Kf5 {4.77/23 5 (e5)} 47. a6 {7.73/28 9 (Db5+)} Bf2 {3.31/19 1} 48. Qb7 {7.86/26 1 (Kc3)} Rxe6 {4.80/20 3} 49. Qh7+ {8.35/22 1} Kf6 {5.08/20 2} 50. Kc3 {35.98/26 5} Re3+ {5.40/21 0 (e3)} 51. Kxc4 {36.00/27 7} Re8 {6.01/22 0 (e6)} 52. Qh6+ {96.05/27 4 (Kc3)} Kf5 {3.27/18 2} 53. Qg7 {199.55/31 8 (Dh7+)} Re4+ {6.12/19 3 (e6)} 54. Kd5 {199.67/48 3 (Kc3)} Rf4 {4.14/23 1} 55. a7 {199.69/73 2} Bxa7 {199.70/42 0} 56. Qf7+ {199.75/69 2} Kg4 {199.72/48 0} 57. Qxa7 {199.79/59 11} Kh5 {199.74/46 0} 58. Qh7+ {199.83/63 21} Kg4 {199.82/49 0} 59. Qh8 {199.85/54 15} Rf2 {199.84/47 0} 60. Qd4+ {199.87/63 11} Rf4 {199.86/43 0} 61. Qd1+ {199.90/59 0} Kg3 {199.90/42 0} 62. Qe2 {199.92/61 4} Rf5+ {199.92/44 0} 63. Ke4 {199.94/58 0 (Ke6)} Rf8 {199.90/40 3 (f4+)} 64. Qe3+ {#30/54 5 (Dd3+)} Kh2 {199.98/29 1 (xg2)} 65. Qxg5 {#20/56 0} Re8+ {2/1 0} 66. Kd5 {1/0 0} Re2 {1/0 0} 67. Qxh4+ {1/1 0} Kxg2 {16/1 0} 68. Kd4 {15/1 0} Kf1 {15/1 0} 69. Kd3 {14/0 0} Rg2 {14/1 0} 70. Qh3 {13/0 0} Kf2 {13/0 0} 71. Qe3+ {12/0 0} Kf1 {12/1 0} 72. Qf3+ {11/1 0} Kg1 {11/1 0} 73. Ke3 {10/1 0} Kh2 {10/0 0} 74. Qd5 {9/1 0} Kh3 {9/1 0} 75. Kf3 {8/1 0} Rg3+ {8/1 0} 76. Kf2 {7/1 0} Kh4 {7/1 0} 77. Qh1+ {6/1 0} Rh3 {6/1 0} 78. Qc1 {5/1 0} Rh2+ {5/0 0} 79. Kf3 {4/1 0} Rh3+ {4/0 0} 80. Kf4 {3/1 0} Rb3 {3/0 0} 81. Qe1+ {2/0 0} Kh5 {2/1 0} 82. Qd1+ {1/1 0} Kg6 {1/0 0} 83. Qxb3 {5/1 0} Kf6 {5/1 0} 84. Qe3 {4/1 0} Kf7 {4/0 0} 85. Kf5 {3/0 0} Kg7 {3/0 0} 86. Qe8 {2/0 0} Kh7 {2/1 0} 87. Kf6 {1/1 0} Kh6 {1/1 0} 88. Qg6# {#1/0 0} 1-0[/pgn]

Such things are not interesting for SF developers, the fish test with bullet games counts, so let's forget that again.

[Modern Times]

And don't forget the unbalanced openings.

[Uri Blass]

Releasing SF 16 on June 30, and then having two NNUE improvements and a code improvement just one day later suggests some disorganization and poor communication within the SF author group. Surely with these big advancements in the pipeline it would have made more sense to wait one day, finalize the tests of the improvements, and only then introduce the latest product as SF 16.

Still, I don't mean to be unduly critical of the SF authors as they do a fine job in crafting this chess engine. However, some disorganization seems evident. So, perhaps they'd benefit from a Team Leader like they once had.

You are effectively arguing that SF development should be stopped for a month after the release of a new version.

No
He does not suggest stopping SF developement but only not releasing a new version when you know that there are improvements that are not in the release.

Stockfish16 is from 29.6 when the following tests ended in 28.6 and (26.6 and 27.6) and they included the two changes only later.

https://tests.stockfishchess.org/tests/ ... ce609d0bff
https://tests.stockfishchess.org/tests/ ... ce609cfd20
https://tests.stockfishchess.org/tests/ ... ce609d0c7d

What is the idea to release stockfish16 in 29.6 when you already know that some tests at LTC passed SPRT and are not included?

It is better to release new official version of stockfish only at a time no new test passed SPRT at LTC

[CornfedForever]

On the bright side...it will end up meaning less of a wait for SF 17!

[smatovic]

This is also a matter of release philosophy in Linux distros, point release vs. rolling release:

https://en.wikipedia.org/wiki/Rolling_release

just grab the newest dev version and compile by yourself?

--
Srdja

[ImNotStockfish]

No
He does not suggest stopping SF developement but only not releasing a new version when you know that there are improvements that are not in the release.

Stockfish16 is from 29.6 when the following tests ended in 28.6 and (26.6 and 27.6) and they included the two changes only later.

https://tests.stockfishchess.org/tests/ ... ce609d0bff
https://tests.stockfishchess.org/tests/ ... ce609cfd20
https://tests.stockfishchess.org/tests/ ... ce609d0c7d

What is the idea to release stockfish16 in 29.6 when you already know that some tests at LTC passed SPRT and are not included?

It is better to release new official version of stockfish only at a time no new test passed SPRT at LTC

Stockfish 16 was released on June 29, but master was frozen on June 22 so all of those changes that you mention couldn't be in SF 16.
Master had to be frozen at some point to make a release, and releases take time to prepare but that doesn't mean that Fishtest is stopped entirely, tests can still be created and they can still pass but they just won't be added to the current release, not a big deal, they will be in the next one.

vondele — 22/06/2023 10:46
so, started a range a PT. I'll consider the branch frozen for a couple of days and see if there are issues coming.
If not, there might be a release near the end of next week.

[Uri Blass]

No
He does not suggest stopping SF developement but only not releasing a new version when you know that there are improvements that are not in the release.

Stockfish16 is from 29.6 when the following tests ended in 28.6 and (26.6 and 27.6) and they included the two changes only later.

https://tests.stockfishchess.org/tests/ ... ce609d0bff
https://tests.stockfishchess.org/tests/ ... ce609cfd20
https://tests.stockfishchess.org/tests/ ... ce609d0c7d

What is the idea to release stockfish16 in 29.6 when you already know that some tests at LTC passed SPRT and are not included?

It is better to release new official version of stockfish only at a time no new test passed SPRT at LTC

Stockfish 16 was released on June 29, but master was frozen on June 22 so all of those changes that you mention couldn't be in SF 16.
Master had to be frozen at some point to make a release, and releases take time to prepare but that doesn't mean that Fishtest is stopped entirely, tests can still be created and they can still pass but they just won't be added to the current release, not a big deal, they will be in the next one.

vondele — 22/06/2023 10:46
so, started a range a PT. I'll consider the branch frozen for a couple of days and see if there are issues coming.
If not, there might be a release near the end of next week.

I understand but I do not like to see official version that has no functional change relative to previous version.
You were afraid that there may be some bug in the version from 22.6 so you released the official version of stockfish 16 only a week later but the result is that practically I will never use the official version because I do not see the point of downloading a new version when there is no functional change relative to the previous developement version that I downloaded.

[abgursu]

No
He does not suggest stopping SF developement but only not releasing a new version when you know that there are improvements that are not in the release.

Stockfish16 is from 29.6 when the following tests ended in 28.6 and (26.6 and 27.6) and they included the two changes only later.

https://tests.stockfishchess.org/tests/ ... ce609d0bff
https://tests.stockfishchess.org/tests/ ... ce609cfd20
https://tests.stockfishchess.org/tests/ ... ce609d0c7d

What is the idea to release stockfish16 in 29.6 when you already know that some tests at LTC passed SPRT and are not included?

It is better to release new official version of stockfish only at a time no new test passed SPRT at LTC

Stockfish 16 was released on June 29, but master was frozen on June 22 so all of those changes that you mention couldn't be in SF 16.
Master had to be frozen at some point to make a release, and releases take time to prepare but that doesn't mean that Fishtest is stopped entirely, tests can still be created and they can still pass but they just won't be added to the current release, not a big deal, they will be in the next one.

vondele — 22/06/2023 10:46
so, started a range a PT. I'll consider the branch frozen for a couple of days and see if there are issues coming.
If not, there might be a release near the end of next week.

I understand but I do not like to see official version that has no functional change relative to previous version.
You were afraid that there may be some bug in the version from 22.6 so you released the official version of stockfish 16 only a week later but the result is that practically I will never use the official version because I do not see the point of downloading a new version when there is no functional change relative to the previous developement version that I downloaded.

I don't think you understand how minority we "who use abrok compiles" are, among the millions of people using Stockfish only via official releases. People simply don't know they exist, and online chess platforms or mobile gui's are no help for that. Lots of people only know when there is an official release and their platforms use the new engine. So, actually, no need to wait or froze master for an official release, abrok users know how to download from there and official-to-official users don't know development versions exist.