Stockfish Rockwood is a monster! :-o

[carldaman]

Recently flying under the radar, SF Rockwood (or simply Stockwood), is a version apparently developed by an entity named 'higgs', somewhere in the middle of the Indian Ocean...

Its play is very impressive, with a more realistic evaluation score than the regular SF, and an ability to find some amazing moves that other versions can't come up with.

[pgn]


[Event "Noomen-pos#11test"]
[Site "?"]
[Date "2014.01.07"]
[Round "1.1"]
[White "Houdini 4 x64A MP c0"]
[Black "Stockwood 030114 x64_sb MP"]
[Result "0-1"]
[ECO "E99"]
[Annotator "0.77;0.47"]
[PlyCount "88"]
[EventDate "2014.01.07"]
[EventType "tourn"]


{Intel(R) Core(TM) i3-2120 CPU @ 3.30GHz 3292 MHz W=20.8 plies; 3,651kN/s; 2
TBAs B=27.5 plies; 2,611kN/s; 1 TBAs} {20moves/10min, 2 cores, 512 MB hash,
3-4-5 man syzygy, ponder off} 1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 5. Nf3 O-O
6. Be2 e5 7. O-O Nc6 8. d5 Ne7 9. Ne1 Nd7 10. Nd3 f5 11. Bd2 Nf6 12. f3 f4 13.
c5 g5 14. cxd6 cxd6 15. Rc1 Ng6 16. Nb5 Rf7 17. Qc2 Ne8 18. a4 h5 19. Nf2 Bd7
$5 {[%eval 47,29] [%emt 0:03:49]} 20. Qd1 $143 {[%eval 77,22] [%emt 0:04:39]} (
20. Na3 $5) (20. Qb3 {most common}) 20... Bf8 $5 $146 {[%eval 36,31] [%emt 0:
06:11] (Nh4) novelty} 21. Kh1 {[%eval 65,23] [%emt 0:04:14]} Rg7 {[%eval 36,29]
[%emt 0:00:24]} 22. Bb4 {[%eval 66,22] [%emt 0:00:22] (b3)} a6 {[%eval 37,25]
[%emt 0:00:33] (Nh4)} 23. Na3 {[%eval 67,22] [%emt 0:00:22]} Nh4 {[%eval 33,25]
[%emt 0:00:25]} 24. Rg1 {[%eval 56,21] [%emt 0:00:37]} Nf6 {[%eval 39,24]
[%emt 0:00:29]} 25. Nc4 {[%eval 47,20] [%emt 0:00:24]} g4 $1 {[%eval 36,26]
[%emt 0:00:44] (Rc8)} 26. Ba5 {[%eval 124,21] [%emt 0:00:23]} Qe8 {[%eval 31,
26] [%emt 0:00:25] (Qe7)} 27. Nb6 {[%eval 101,21] [%emt 0:01:21]} g3 $1 $40 {
[%eval 33,27] [%emt 0:00:29] Black ignores White's Q-side forays, with good
reason, typical of the classical KID} 28. Nxd7 {[%eval 88,21] [%emt 0:00:49]} (
28. Nxa8 $2 gxf2 29. Nc7 fxg1=Q+ {loses a piece}) 28... Rxd7 $1 {[%eval 36,27]
[%emt 0:00:59]} ({less good is} 28... Nxd7 $143 29. hxg3 fxg3 30. Nh3 Nc5 (
30... Nf6 31. Bd2 $1) 31. b3 $14 (31. Rxc5 $5 $44)) 29. Nd3 $143 $6 {[%eval 86,
21] [%emt 0:00:25] (Nh3) this was a key moment} (29. hxg3 fxg3 30. Nh3 $8 (30.
Nd3 $2 Qe7 $1 $19 31. Be1 Nf5 $3 32. exf5 Ne4 $19 33. -- $140 Qh4#) 30... Bh6
$13 {reason why 28...Rxd7 was preferred}) (29. Nh3 $5 {this needs to be tried})
29... Rg7 {[%eval 24,25] [%emt 0:00:18] (gxh2)} 30. Be1 {[%eval 65,20] [%emt 0:
01:55] (Rc7)} Qg6 {[%eval 16,21] [%emt 0:00:24] (Rc8)} 31. a5 {[%eval 60,21]
[%emt 0:01:15] (b3)} Ng4 $3 $132 {[%eval -10,28] [%emt 0:00:41] (gxh2) latest
SF dev version cannot find this move} 32. h3 {[%eval 24,21] [%emt 0:00:54]
(fxg4) now this pawn becomes a target} Ne3 {[%eval 4,30] [%emt 0:00:36]} 33.
Qa4 {[%eval 24,19] [%emt 0:00:00]} Qg5 {[%eval 6,30] [%emt 0:00:25] (Qe8)} 34.
Bd2 {[%eval 45,20] [%emt 0:00:14]} ({threat was} 34. -- Nhxg2 $1 35. Rxg2 Qh4
$19) 34... Nexg2 $1 {[%eval 4,31] [%emt 0:00:17]} ({now} 34... Nhxg2 $2 35.
Rxg2 Qh4 36. Bxe3) 35. Rxg2 {[%eval 45,18] [%emt 0:00:00]} Qd8 $5 {[%eval 1,31]
[%emt 0:00:23] (Rd8) Rockwood has a diabolical plan in mind!} 36. Rgg1 {[%eval
26,21] [%emt 0:00:39] (Ne1)} Rc8 $1 {[%eval 0,27] [%emt 0:00:26]} 37. Rgd1 {
[%eval -3,22] [%emt 0:00:30] (Qd1)} Rc6 $3 {[%eval -295,28] [%emt 0:00:19]
(Rcc7) one of the most spectacular moves I've seen in a long time} 38. Rg1 {
[%eval -548,17] [%emt 0:00:22]} (38. dxc6 $140 Qc8 $1 $19 {taking over this
diagonal was the idea behind Qd8/Rc8/Rc6 maneuver}) 38... g2+ $5 {[%eval -341,
28] [%emt 0:00:17] totally winning} ({of course, also decisive is} 38... Qc8
39. Bf1 Rxc1 40. Bxc1 Nxf3 41. Bg2 Nxg1 42. Kxg1 Rf7 $1 43. Qd1 f3 $1 $19) 39.
Kh2 {[%eval -548,16] [%emt 0:00:00]} Qg5 {[%eval -408,29] [%emt 0:00:32]
Houdini is getting crushed} 40. Rge1 {[%eval -612,24] [%emt 0:00:35] prevents
one mate, but...} Rxc1 {[%eval -452,29] [%emt 0:00:48]} 41. Kg1 {[%eval -589,
22] [%emt 0:00:00]} (41. Rxc1 $4 g1=Q+) 41... Rc8 {[%eval -451,29] [%emt 0:00:
21] (Rcc7)} 42. Nf2 {[%eval -742,22] [%emt 0:01:47] (b4)} Qg3 {[%eval -528,28]
[%emt 0:00:36]} 43. Bc3 {[%eval -808,22] [%emt 0:01:13] (Bd1)} Be7 {[%eval
-711,25] [%emt 0:00:20] (Nxf3+)} 44. Qc2 {[%eval -1052,22] [%emt 0:01:36] (Nd3)
} Bd8 {[%eval -934,26] [%emt 0:00:33] (Kh7) adj 0-1 : Black is up the exchange
and dominates both sides of the board} ({play could continue} 44... Bd8 45. Qd3
Nf5 $3 46. exf5 Rxc3 $1 (46... Bh4 $19 47. Bd4 exd4 48. Qxd4 Re8 $19) 47. Qxc3
Bh4 $19 {and mate to follow}) 0-1

[/pgn]

Nice game, eh? Former king-of-the-hill Houdini was made to look pedestrian. A couple of highlights below:

[D]r4bk1/1p4r1/p2p2q1/P2Pp2p/4Ppnn/3N1Pp1/1P2B1PP/2RQB1RK w - - 1 32

Stockwood has just played Ng4! This is the kind of chess I like, especially in the KID

In the next diagram, can you guess the right move?


[D]2rq1bk1/1p4r1/p2p4/P2Pp2p/Q3Pp1n/3N1PpP/1P1BB3/2RR3K b - - 4 37

(answer in game notes)

Best regards,
CL

[syzygy]

Recently flying under the radar, SF Rockwood (or simply Stockwood), is a version apparently developed by an entity named 'higgs', somewhere in the middle of the Indian Ocean...

This is supposed to be the link to the sources.
The archive opens, but is password protected.
GPL violation. Illegal.

Anyway, as far as I understand this is just Stockfish with large page support patched in. It won't find moves that SF can't find.

[Mike S.]

Very attractive examples! I cannot judge which SF. versions, or other engines can find one move or the other. Just from a quick try:

(dual i5, 4 threads, 512 MB hash)

Stockfish DD PG (= recent PA_GTB version)
1/1 00:00 198 33k -1,10 31. ... gxh2 32.Bxh4 hxg1Q+ 33.Qxg1
(...)
29/49+ 02:23 418 369k 2 911k +0,59 31. ... Re8 32.hxg3 fxg3 33.Rc4 Nd7 34.Rf1 Qf6 35.Kg1 Ree7 36.Qc1 Ref7 37.Bd1 Be7 38.Qd2 b5 39.Rc8+ Rf8 40.Rxf8+ Nxf8 41.f4
29/49- 02:48 497 000k 2 943k +0,35 31. ... Ng4 32.hxg3 fxg3 33.Qd2 Nh6 34.Rf1 Qf6 35.f4 N4f5
(...)
33/49+ 06:03 1 085 118k 2 983k +0,24 31. ... Ng4 32.h3 Ne3 33.Qa4 Be7 34.Bd2 Nhxg2 35.Rxg2 Nxg2 36.Kxg2 Bh4 37.Qb4 Kh8 38.Qb6 Bd8 39.Qb3 Bh4 40.Be1 Qf7 41.Qb6 Qd7 42.Bb4 Rag8
Best move: Ng4, Value: +0,24, Depth: 33/49, Time: 06:03,761, 1 085 117 649 Nodes, 2 983k

The other move 37...Rc6!! is indeed spectacular! - But maybe less difficult for engines:

Stockfish DD PG:
1/2 00:00 162 32k +1,49 37. ... Be7
(...)
21/28- 00:04 5 939k 1 384k -0,28 37. ... Rb8 38.Rc3 b5 39.axb6 Qxb6 40.Rg1 Qa7 41.Be1 Kh7 42.Qc6 a5 43.Bd2 Rxb2 44.Nxb2 Qf2
21/32- 00:05 9 518k 1 664k -0,46 37. ... Rc6 38.dxc6 Qc8 39.Qc4+ Kh8 40.Bf1 g2+ 41.Bxg2 Rxg2 42.Nxf4 exf4 43.Bc3+ Bg7 44.Bxg7+ Kxg7 45.Qd4+ Kh7
(...)
31/56 01:09 222 385k 3 182k -8,76 37. ... Rc6 38.Rg1 g2+ 39.Kh2 Qg5 40.Rge1 Rxc1 41.Kg1 Rc8 42.Rb1 Qg3 43.Qd1 Qxh3 44.Nf2 Qg3 45.Bb4 Be7 46.Ra1 Rc7 47.Rb1 Rg6 48.Bd2 Bd8 49.b4 Rcg7 50.Rc1 Bc7 51.Qe1 Kh7 52.Nd3 Qh3 53.Nf2 Qg3 54.Nd3 Qh3 55.Nf2 Qg3
Best move: Rc6, Value: -8,76, Depth: 31/56, Time: 01:09,901, 222 385 323 Nodes, 3 182k

Rybka 2.3.2a:
5 00:00 3k 208k +0,85 37. ... b5 38.axb6
(...)
18 00:27 8 916k 331k 0,00 37. ... Rcc7 38.Rxc7 Rxc7 39.Rc1 Qc8 40.Bf1 Nxf3 41.Rxc7 Qxc7 42.Qd1 Nxd2 43.Qxd2 f3 44.b4
18 01:26 29 026k 344k -2,09 37. ... Rc6 38.dxc6 Qc8 39.Qb3+ Kh8 40.Bf1 g2+ 41.Bxg2 Rxg2 42.Nxf4 exf4 43.Bc3+ Bg7 44.Bxg7+
18 03:10 65 167k 350k -2,66 37. ... Rc6 38.Rg1 g2+ 39.Kh2 Qg5 40.Rgd1 Rxc1 41.Kg1 Rxd1+ 42.Qxd1 Qg3 43.Nf2 Rc7 44.Be1
Best move: Rc6, Value: -2,66, Depth: 19, Time: 03:15,261, 66 702 540 Nodes, 350k

[carldaman]

What I meant was that SF DD could not find the speculative 31...Ng4 at the same time control -- I analyze with it (PA_GTB) all the time. With more time it will see it, of course.

And Houdini did not see 37...Rc6 coming within the time control setting (it was expecting Rcc7). Let's not forget that Houdini had been the tactical standard for a long time, but now it is lagging behind SF. This is very impressive in my book.

[carldaman]

Recently flying under the radar, SF Rockwood (or simply Stockwood), is a version apparently developed by an entity named 'higgs', somewhere in the middle of the Indian Ocean...

This is supposed to be the link to the sources.
The archive opens, but is password protected.
GPL violation. Illegal.

Anyway, as far as I understand this is just Stockfish with large page support patched in. It won't find moves that SF can't find.

Hmm... why would he pw-protect the sources?...

Re: your second point -- it does find certain key moves faster than regular SF.
I did not use large pages, since I noticed that it significantly lowered nps.

[ernst]

Recently flying under the radar, SF Rockwood (or simply Stockwood), is a version apparently developed by an entity named 'higgs', somewhere in the middle of the Indian Ocean...

This is supposed to be the link to the sources.
The archive opens, but is password protected.
GPL violation. Illegal.

Anyway, as far as I understand this is just Stockfish with large page support patched in. It won't find moves that SF can't find.

The source is also included in the [LINK DELETED by moderation] download.

[MODERATION EDIT]
The package contains a link to the sources, which are password protected. This is highly irregular for a GPL program. If the "author" does not provide that password, it is a clear violation of the license.

[syzygy]

The source is also included in the [LINK DELETED by moderation] download.

Did you actually look?

[notyetagm]

Is Stockfish DD PG the most recent stable version? If not, how experimental is it? And what does the "PG" stand for?

Thanks

Very attractive examples! I cannot judge which SF. versions, or other engines can find one move or the other. Just from a quick try:

(dual i5, 4 threads, 512 MB hash)

Stockfish DD PG (= recent PA_GTB version)
1/1 00:00 198 33k -1,10 31. ... gxh2 32.Bxh4 hxg1Q+ 33.Qxg1
(...)
29/49+ 02:23 418 369k 2 911k +0,59 31. ... Re8 32.hxg3 fxg3 33.Rc4 Nd7 34.Rf1 Qf6 35.Kg1 Ree7 36.Qc1 Ref7 37.Bd1 Be7 38.Qd2 b5 39.Rc8+ Rf8 40.Rxf8+ Nxf8 41.f4
29/49- 02:48 497 000k 2 943k +0,35 31. ... Ng4 32.hxg3 fxg3 33.Qd2 Nh6 34.Rf1 Qf6 35.f4 N4f5
(...)
33/49+ 06:03 1 085 118k 2 983k +0,24 31. ... Ng4 32.h3 Ne3 33.Qa4 Be7 34.Bd2 Nhxg2 35.Rxg2 Nxg2 36.Kxg2 Bh4 37.Qb4 Kh8 38.Qb6 Bd8 39.Qb3 Bh4 40.Be1 Qf7 41.Qb6 Qd7 42.Bb4 Rag8
Best move: Ng4, Value: +0,24, Depth: 33/49, Time: 06:03,761, 1 085 117 649 Nodes, 2 983k

The other move 37...Rc6!! is indeed spectacular! - But maybe less difficult for engines:

Stockfish DD PG:
1/2 00:00 162 32k +1,49 37. ... Be7
(...)
21/28- 00:04 5 939k 1 384k -0,28 37. ... Rb8 38.Rc3 b5 39.axb6 Qxb6 40.Rg1 Qa7 41.Be1 Kh7 42.Qc6 a5 43.Bd2 Rxb2 44.Nxb2 Qf2
21/32- 00:05 9 518k 1 664k -0,46 37. ... Rc6 38.dxc6 Qc8 39.Qc4+ Kh8 40.Bf1 g2+ 41.Bxg2 Rxg2 42.Nxf4 exf4 43.Bc3+ Bg7 44.Bxg7+ Kxg7 45.Qd4+ Kh7
(...)
31/56 01:09 222 385k 3 182k -8,76 37. ... Rc6 38.Rg1 g2+ 39.Kh2 Qg5 40.Rge1 Rxc1 41.Kg1 Rc8 42.Rb1 Qg3 43.Qd1 Qxh3 44.Nf2 Qg3 45.Bb4 Be7 46.Ra1 Rc7 47.Rb1 Rg6 48.Bd2 Bd8 49.b4 Rcg7 50.Rc1 Bc7 51.Qe1 Kh7 52.Nd3 Qh3 53.Nf2 Qg3 54.Nd3 Qh3 55.Nf2 Qg3
Best move: Rc6, Value: -8,76, Depth: 31/56, Time: 01:09,901, 222 385 323 Nodes, 3 182k

Rybka 2.3.2a:
5 00:00 3k 208k +0,85 37. ... b5 38.axb6
(...)
18 00:27 8 916k 331k 0,00 37. ... Rcc7 38.Rxc7 Rxc7 39.Rc1 Qc8 40.Bf1 Nxf3 41.Rxc7 Qxc7 42.Qd1 Nxd2 43.Qxd2 f3 44.b4
18 01:26 29 026k 344k -2,09 37. ... Rc6 38.dxc6 Qc8 39.Qb3+ Kh8 40.Bf1 g2+ 41.Bxg2 Rxg2 42.Nxf4 exf4 43.Bc3+ Bg7 44.Bxg7+
18 03:10 65 167k 350k -2,66 37. ... Rc6 38.Rg1 g2+ 39.Kh2 Qg5 40.Rgd1 Rxc1 41.Kg1 Rxd1+ 42.Qxd1 Qg3 43.Nf2 Rc7 44.Be1
Best move: Rc6, Value: -2,66, Depth: 19, Time: 03:15,261, 66 702 540 Nodes, 350k

[ernst]

The source is also included in the [LINK DELETED by moderation] download.

Did you actually look?

Ah I see, the upload has been replaced. Still have the source here though as I downloaded it earlier when this was included.

[Mike S.]

Stockfish DD PG = Stockfish PA_GTB DD 64 SSE4.2 (I have just shortened the name). This is a branch being done by Jeremy Bernstein, who adds Gaviota tbs. access and permanent learning to SF. (among other changes), in this case to Stockfish DD:

http://open-chess.org/viewtopic.php?f=7 ... 910#p19910

Runs very fine here.

The official SF. DD is at http://blog.stockfishchess.org/ and is currently the latest major release, but I consider the dev. versions from http://abrok.eu/stockfish/ as stable too, as they are always tested with many thousands of games. I am still using the versions of Dec. 10th, 2013 and Dec. 13th, 2013 (Syz.) respectively, because I cannot keep up with their developement speed.