Deep Blue

[darmar]

IBM Team has released new gun:
Moderation: Link removed. Likely a copyright violation or a virus.
First tests +70 Elo ahead of Stockfish

[Uri Blass]

I do not believe it and do not download a file that may be a virus.

[Chacal X]

IBM Team has released new gun:
Moderation: Link removed. Likely a copyright violation or a virus.
First tests +70 Elo ahead of Stockfish

Excellent engine, congratulations to the programmers.
Flavio/

[peter]

Output looks rather SF- like.
Here's the already well known Salai- study once in a while again
https://www.yacpdb.org/#search/NEsxazFc ... LzEvMA==/1

Code: Select all

Deep Blue by IBM Team
position fen 4K1k1/8/1p5p/1Pp3b1/8/1P3P2/P1B2P2/8 w - -
setoption name hash value 1024
go depth 60
info string Available processors: 0-31
info string Using 1 thread
info string NNUE evaluation using nn-killer.nnue (125MiB, (102384, 1024, 15, 32, 1))
info string NNUE evaluation using nn-assassin.nnue (6MiB, (22528, 128, 15, 32, 1))
info string Network replica 1: Shared memory.
info depth 1 seldepth 2 multipv 1 score cp 181 nodes 15 nps 15000 hashfull 0 tbhits 0 time 1 pv e8d7
...
info depth 60 seldepth 90 multipv 1 score cp 28 nodes 313229697 nps 1758719 hashfull 504 tbhits 0 time 178101 pv e8d7 g5d2 d7e6 g8g7 c2d3 h6h5 d3f1 h5h4 a2a4 d2a5 f3f4 a5c3 f4f5 g7f8 f5f6 c3e1 f2f3 e1c3 f3f4 f8e8 f1g2 e8f8 e6f5 f8f7 g2d5 f7f8 f5g4 c3f6 a4a5 b6a5 b5b6 f6d8 b6b7 d8c7 f4f5 c7g3 f5f6 g3e5 d5e6 e5g3 e6c4 f8e8 g4f5 g3b8 f6f7 e8f8 f5g4 b8g3 g4g5 f8g7 g5f5 g3b8 f5g4 b8g3 c4e6 g7f8 g4g5 f8e7 e6c4 e7f8 c4d5 f8e7 g5f5 g3b8 f5g4 b8g3 g4h3 e7f8 d5e6
bestmove e8d7 ponder g5d2

info string Network replica 1: Shared memory.

sounds a little like "verbatim net patch" doesn't it? That's to be read in SF TI too, SF dev. with this one patch only I haven't downloaded so far.

Anyhow 1.f4 is not found in depth 60, yet with 30 threads and MultiPV=10 <1 minute. That's some better then with SF dev 20251014, even if there's lot of time to solution- spreading is such a positions with 30 threads.

SF 20251014:

Code: Select all

Stockfish dev-20251014-75edbee0 by the Stockfish developers (see AUTHORS file)
position fen 4K1k1/8/1p5p/1Pp3b1/8/1P3P2/P1B2P2/8 w - -
setoption name hash value 1024
go depth 60
info string Available processors: 0-31
info string Using 1 thread
info string NNUE evaluation using nn-1c0000000000.nnue (133MiB, (22528, 3072, 15, 32, 1))
info string NNUE evaluation using nn-37f18f62d772.nnue (6MiB, (22528, 128, 15, 32, 1))
info depth 1 seldepth 2 multipv 1 score cp 220 nodes 14 nps 14000 hashfull 0 tbhits 0 time 1 pv e8d7
...
info depth 60 seldepth 98 multipv 1 score cp 38 nodes 358356114 nps 1774655 hashfull 571 tbhits 0 time 201930 pv e8d7 g5d2 d7e6 g8g7 c2d3 h6h5 e6f5 h5h4 f5g4 g7f6 g4h4 d2f4 h4g4 f6e5 d3g6 f4c1 g4g3 c1d2 g6d3 e5d4 d3f1 d4e5 g3g4 d2f4 f1c4 f4c1 g4g3 c1d2 g3g2 e5f4 c4d5 d2c3 d5e4 c3a5 g2f1 a5c3 f1e2 c3a5 e4g6 a5c3 g6f7 c3a5 f7e6 a5b4 e6h3 b4c3 h3g2 c3d4 e2e1 d4c3 e1d1 c3a5 d1e2 a5b4 e2d3 b4e1
bestmove e8d7 ponder g5d2

So best move again isn't found in depth 60 with single thread.

If it's SF, then probably not the TI- one, looking like this:

Code: Select all

Stockfish dev-20251109-nogit by the Stockfish developers (see AUTHORS file)
position fen 4K1k1/8/1p5p/1Pp3b1/8/1P3P2/P1B2P2/8 w - -
setoption name hash value 1024
go depth 60
info string Available processors: 0-31
info string Using 1 thread
info string NNUE evaluation using nn-49c1193b131c.nnue (125MiB, (102384, 1024, 15, 32, 1))
info string NNUE evaluation using nn-37f18f62d772.nnue (6MiB, (22528, 128, 15, 32, 1))
info string Network replica 1: Shared memory.
info depth 1 seldepth 2 multipv 1 score cp 181 nodes 15 nps 15000 hashfull 0 tbhits 0 time 1 pv e8d7
---
info depth 53 seldepth 90 multipv 1 score cp 52 nodes 128348320 nps 1818067 hashfull 254 tbhits 0 time 70596 pv e8d7 g5d2 d7e6 g8g7 a2a4 h6h5 e6f5 h5h4 f5g4 g7f7 f3f4 f7f6 g4h4 d2a5 c2d3 a5b4 d3c4 f6f5 h4g3 b4a5 g3f3 f5f6 f3e4 a5b4 f4f5 b4a5 e4f3 f6f5 c4d3 f5e5 f3g4 a5d2 d3c2 d2a5 f2f4 e5e6 g4g5 a5d2 c2e4 e6e7 g5f5 e7f7 e4d3 f7e7 d3e2 e7d7 f5e4 d7e7 e2f1 e7f6
...
info depth 54 seldepth 64 multipv 1 score cp 58 upperbound nodes 131005132 nps 1820552 hashfull 260 tbhits 0 time 71959 pv e8d7 g5d2
info depth 54 currmove e8d7 currmovenumber 1
info depth 54 currmove f3f4 currmovenumber 2
info depth 54[/b] seldepth 77 multipv 1 score cp 58 lowerbound nodes 138848873 nps 1829896 hashfull 272 tbhits 0 time 75878 pv f3f4
info depth 53 currmove f3f4 currmovenumber 1
info depth 54 seldepth 77 multipv 1 score cp 71 lowerbound nodes 138976721 nps 1829869 hashfull 272 tbhits 0 time 75949 pv f3f4
...
info depth 55 seldepth 85 multipv 1 score cp 721 nodes 511230695 nps 1864675 hashfull 906 tbhits 0 time 274166 pv f3f4 g5f6 a2a4 f6c3 f4f5 g8g7 e8e7 c3f6 e7e6 f6c3 c2d1 g7f8 f5f6 c3a5 f6f7 a5e1 d1h5 e1f2 a4a5 b6a5 e6d5 f8e7 b5b6 c5c4 b6b7 f2g3 d5c4 g3c7 c4c5 c7h2 c5b5 h2b8 b5a5 b8a7 b3b4 a7b8 a5b6 e7f8 b6c6 f8e7 f7f8q e7f8 c6d7 b8g3 d7c8 f8e7 b7b8q g3b8 c8b8 e7f6 b4b5 f6g5 h5d1 h6h5 b8c7
info depth 56 currmove f3f4 currmovenumber 1
stop
info depth 56 seldepth 85 multipv 1 score cp 721 nodes 538715842 nps 1876705 hashfull 921 tbhits 0 time 287054 pv f3f4 g5f6 a2a4 f6c3 f4f5 g8g7 e8e7 c3f6 e7e6 f6c3 c2d1 g7f8 f5f6 c3a5 f6f7 a5e1 d1h5 e1f2 a4a5 b6a5 e6d5 f8e7 b5b6 c5c4 b6b7 f2g3 d5c4 g3c7 c4c5 c7h2 c5b5 h2b8 b5a5 b8a7 b3b4 a7b8 a5b6 e7f8 b6c6 f8e7 f7f8q e7f8 c6d7 b8g3 d7c8 f8e7 b7b8q g3b8 c8b8 e7f6 b4b5 f6g5 h5d1 h6h5 b8c7
bestmove f3f4 ponder g5f6

info string Network replica 1: Shared memory

again.

Yet SF TI finds 1.f4 in about 76 seconds, a very fine performance single threaded (3.5GHz).

SMP benchmarks compared, first one again "Deep Blue":

Code: Select all

Deep Blue by IBM Team
bench 8192 30 19
info string Using 30 threads

Position: 1/48 (rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1)
info string NNUE evaluation using nn-killer.nnue (125MiB, (102384, 1024, 15, 32, 1))
info string NNUE evaluation using nn-assassin.nnue (6MiB, (22528, 128, 15, 32, 1))
info string Network replica 1: Shared memory.
info depth 1 seldepth 2 multipv 1 score cp 3 nodes 2210 nps 736666 hashfull 0 tbhits 0 time 3 pv d2d4
...
Total time (ms) : 89738
Nodes searched  : 2589558954
Nodes/second    : 28856882

and SF 20251014:

Code: Select all

Stockfish dev-20251014-75edbee0 by the Stockfish developers (see AUTHORS file)
bench 8192 30 19
info string Using 30 threads

Position: 1/48 (rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1)
info string NNUE evaluation using nn-1c0000000000.nnue (133MiB, (22528, 3072, 15, 32, 1))
info string NNUE evaluation using nn-37f18f62d772.nnue (6MiB, (22528, 128, 15, 32, 1))
info depth 1 seldepth 2 multipv 1 score cp 34 nodes 2267 nps 1133500 hashfull 0 tbhits 0 time 2 pv e2e4
...
Total time (ms) : 121141
Nodes searched  : 3167649469
Nodes/second    : 26148450

One more similarity to SF: in Fritz 20 automatically running test- suite get's frozen about 3 times per 100 positions as well with SF TI as with "Deep Blue", regards

[peter]

A match against Stockfish TI with 3'+1" TC, single thread, UHO 2024 900-990cp 6mvs, stopped after 273 games because of disconnection of Stockfish TI in CuteChess GUI:

Code: Select all

Score of StockfishTI251109 vs DeepBlue: 72 - 49 - 124 [0.547]
...      StockfishTI251109 playing White: 71 - 2 - 51  [0.778] 124
...      StockfishTI251109 playing Black: 1 - 47 - 73  [0.310] 121
...      White vs Black: 118 - 3 - 124  [0.735] 245
Elo difference: 32.7 +/- 30.6, LOS: 98.2 %, DrawRatio: 50.6 %
SPRT: llr 0 (0.0%), lbound -inf, ubound inf
273 of 500 games finished.

Player: StockfishTI251109
   "Draw by 3-fold repetition": 8
   "Draw by adjudication": 80
   "Draw by adjudication: SyzygyTB": 36
   "Loss: Black wins by adjudication: SyzygyTB": 1
   "Loss: White mates": 10
   "Loss: White wins by adjudication: SyzygyTB": 37
   "Loss: White's connection stalls": 1
   "No result": 28
   "Win: White mates": 18
   "Win: White wins by adjudication: SyzygyTB": 53
   "Win: White's connection stalls": 1
Player: DeepBlue
   "Draw by 3-fold repetition": 8
   "Draw by adjudication": 80
   "Draw by adjudication: SyzygyTB": 36
   "Loss: White mates": 18
   "Loss: White wins by adjudication: SyzygyTB": 53
   "Loss: White's connection stalls": 1
   "No result": 28
   "Win: Black wins by adjudication: SyzygyTB": 1
   "Win: White mates": 10
   "Win: White wins by adjudication: SyzygyTB": 37
   "Win: White's connection stalls": 1

Here's the position (opening line) SF stopped at:
[pgn][Event "SFTI-DeBl"]
[Site "?"]
[Date "2025.11.11"]
[Round "136"]
[White "StockfishTI251109"]
[Black "DeepBlue"]
[Result "0-1"]
[ECO "B20"]
[PlyCount "12"]
[GameDuration "00:00:15"]
[GameEndTime "2025-11-11T14:46:32.181 Mitteleuropäische Zeit"]
[GameStartTime "2025-11-11T14:46:17.153 Mitteleuropäische Zeit"]
[Opening "Sicilian defense"]
[Termination "stalled connection"]
[TimeControl "180+1"]

1.e4 {book} 1...c5 {book} 2.b3 {book} 2...b6 {book} 3.Bb2 {book} 3...Bb7 {book} 4.Nc3 {book} 4...d6 {book} 5.f4 {book} 5...e6 {book} 6.Nf3 {book} 6...Nd7 {book, White's connection stalls} 0-1
[/pgn]
With same line in reversed colour- game White's connection stalls with Deep Blue playing with White also, maybe because SF had already disconnected, maybe independently, had no log- file set, anyhow the win for disconnection is counted by GUI for Deep Blue only.
In Fritz- GUI with single trial the crash of SF isn't reproducible, regards

[Joerg Oster]

Of course, this is Stockfish!
And according to the name and the reported dimensions of the nets, it is Stockfish-TI ...
maybe tweaked or not, but who really cares?

(And yes, it is a violation of the GPL by not including the sources or a link to them.)

[Ras]

Moderation: If the IBM team had legitimately released an engine, it wouldn't be on some sketchy share-hoster.

[chessica]

Moderation: If the IBM team had legitimately released an engine, it wouldn't be on some sketchy share-hoster.

right!

[Ras]

Moderation: Removed some posts. No need to attack each other. Please use the reporting button if need be, that's it.

[Werewolf]

Moderation: If the IBM team had legitimately released an engine, it wouldn't be on some sketchy share-hoster.

^ this