Surprise result 16 cores from CSS forum

Discussion of computer chess matches and engine tournaments.

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Jouni
Posts: 2334
Joined: Wed Mar 08, 2006 7:15 pm

Surprise result 16 cores from CSS forum

Post by Jouni » Sat Dec 26, 2015 8:04 pm

Hardware: Dual AMD Opteron 6376 @ 2.6 GHz
Bedingungen: Cutechess-Cli, je Engine 16 Kerne, ponder off, 8192 MB Hash, 5-Men Syzygy Bases, 120 Minuten + 30 Sekunden, Komodo Contempt=0; other settings default

Code: Select all

1 Stockfish 15112502 T16  : 3009   11   11   132    52.7 %   2991   85.6 %
2 Komodo 9.3 T16          : 2991   11   11   132    47.3 %   3009   85.6 %

Jouni

beram
Posts: 1187
Joined: Wed Jan 06, 2010 2:11 pm

Re: Surprise result 16 cores from CSS forum

Post by beram » Sat Dec 26, 2015 10:36 pm

Jouni wrote:Hardware: Dual AMD Opteron 6376 @ 2.6 GHz
Bedingungen: Cutechess-Cli, je Engine 16 Kerne, ponder off, 8192 MB Hash, 5-Men Syzygy Bases, 120 Minuten + 30 Sekunden, Komodo Contempt=0; other settings default

Code: Select all

1 Stockfish 15112502 T16  : 3009   11   11   132    52.7 %   2991   85.6 %
2 Komodo 9.3 T16          : 2991   11   11   132    47.3 %   3009   85.6 %

These two positions arose in two of the four games won with black by SF in these match on 16 cores played by Andreas Strangmuller on his 16 core system at 120m30s avg depth 30 ply in early middlegame
In both positions Komodo had his eval at 0.00 at 30ply and 34ply before the move played

32. ..g5! starting a kingside attack and within 10 moves +4 for SF
[d] rr1q2k1/3b2bp/3p2p1/3Pp3/p1P1Pp2/Rp3P1P/1P1NQBP1/R5K1 b - - 0 32
And this one 37..Rf3 ! starting a fierce attack on black king
[d] 1r1q1r2/6pk/3p2b1/p2Nn2p/2B1p2b/1PP4P/R3QPP1/3N1RK1 b - - 0 37

carldaman
Posts: 2048
Joined: Sat Jun 02, 2012 12:13 am

Re: Surprise result 16 cores from CSS forum

Post by carldaman » Sun Dec 27, 2015 12:31 am

beram wrote:
Jouni wrote:Hardware: Dual AMD Opteron 6376 @ 2.6 GHz
Bedingungen: Cutechess-Cli, je Engine 16 Kerne, ponder off, 8192 MB Hash, 5-Men Syzygy Bases, 120 Minuten + 30 Sekunden, Komodo Contempt=0; other settings default

Code: Select all

1 Stockfish 15112502 T16  : 3009   11   11   132    52.7 %   2991   85.6 %
2 Komodo 9.3 T16          : 2991   11   11   132    47.3 %   3009   85.6 %

These two positions arose in two of the four games won with black by SF in these match on 16 cores played by Andreas Strangmuller on his 16 core system at 120m30s avg depth 30 ply in early middlegame
In both positions Komodo had his eval at 0.00 at 30ply and 34ply before the move played

32. ..g5! starting a kingside attack and within 10 moves +4 for SF
[d] rr1q2k1/3b2bp/3p2p1/3Pp3/p1P1Pp2/Rp3P1P/1P1NQBP1/R5K1 b - - 0 32
And this one 37..Rf3 ! starting a fierce attack on black king
[d] 1r1q1r2/6pk/3p2b1/p2Nn2p/2B1p2b/1PP4P/R3QPP1/3N1RK1 b - - 0 37
Hi Bram,
are these games downloadable from anywhere?

Thanks,
CL

User avatar
Laskos
Posts: 10948
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: Surprise result 16 cores from CSS forum

Post by Laskos » Sun Dec 27, 2015 1:59 am

carldaman wrote:
Hi Bram,
are these games downloadable from anywhere?

Thanks,
CL
http://www.fastgm.de/schach/K93-SF15112 ... 200+30.pgn

APassionForCriminalJustic
Posts: 416
Joined: Sat May 24, 2014 7:16 am

Re: Surprise result 16 cores from CSS forum

Post by APassionForCriminalJustic » Sun Dec 27, 2015 9:47 pm

Jouni wrote:Hardware: Dual AMD Opteron 6376 @ 2.6 GHz
Bedingungen: Cutechess-Cli, je Engine 16 Kerne, ponder off, 8192 MB Hash, 5-Men Syzygy Bases, 120 Minuten + 30 Sekunden, Komodo Contempt=0; other settings default

Code: Select all

1 Stockfish 15112502 T16  : 3009   11   11   132    52.7 %   2991   85.6 %
2 Komodo 9.3 T16          : 2991   11   11   132    47.3 %   3009   85.6 %

Change the thread count to 32 then let's see what happens. You're using an Opteron?! Come on man. Your ram (hash) is lacking, your table bases are lacking, and the time control is still short of that which was used at TCEC.

Take the results with a grain of salt. We need results with real good Intel dual socket systems. AMD is garbage. Komodo is 100 percent developed with Intel CPUs.

carldaman
Posts: 2048
Joined: Sat Jun 02, 2012 12:13 am

Re: Surprise result 16 cores from CSS forum

Post by carldaman » Mon Dec 28, 2015 1:36 am

Laskos wrote:
carldaman wrote:
Hi Bram,
are these games downloadable from anywhere?

Thanks,
CL
http://www.fastgm.de/schach/K93-SF15112 ... 200+30.pgn
Thanks Kai, I'd looked at fastgm.de previously, but could not see it.

Regards,
CL

User avatar
Graham Banks
Posts: 34847
Joined: Sun Feb 26, 2006 9:52 am
Location: Auckland, NZ

Re: Surprise result 16 cores from CSS forum

Post by Graham Banks » Mon Dec 28, 2015 2:27 am

APassionForCriminalJustic wrote:.......AMD is garbage. Komodo is 100 percent developed with Intel CPUs.
I don't think that statement is fair on two fronts, firstly because it is a big putdown to those using AMD and secondly because engine authors should realise that many engine users will use AMD architecture, just as some have more modern computers than others.

I think it has been proven previously that Komodo and other engines don't suffer through running on AMD?
gbanksnz at gmail.com

beram
Posts: 1187
Joined: Wed Jan 06, 2010 2:11 pm

Re: Surprise result 16 cores from CSS forum

Post by beram » Mon Dec 28, 2015 10:12 pm

Game comments on the 16core match between SF 251115 and K93

Overall SF wins 4 times with black and 9 times with white
Komodo wins 6 times with white
All four won black games are worth playing trough.
My brief comments are on the won white games trying to make visible the differences in the way the engines won their games

SF white wins:

Game 4 a gradually outplayed win right out a classical Caro Kann 4..Bf5 mainline. SF holds on to a small advantage having better pawn structure. Around move 62 the advantage became +1 (technically endgame win, converted from out of the opening through the middlegame)
[d] 4R3/p1kr2p1/2p2p1p/7P/Pn2K1P1/1P3P2/8/2B5 b - - 0 62

Game 32 a E92 classical King Indian. Black bishop from g7 has made a strange route by d4 e3 to a7 and points at the white king, but leaving its own king thus unprotected
[d] r3q3/bppb1pkn/n2p2r1/3Pp3/1PP1P3/2NB1N2/4Q1Pp/1R2BR1K b - - 0 29
After Be1 white threatens to invade with knight to f5, Komodo prevents that by playing Bg4 but is forced to exchange after the advantage rises quickly for white.
Black gets temporarily 6 pawns for 4 but its king is exposed and white has upperhand on queenside and continues by combing those threats for a win
(a positionally win by outplaying the opponent)
[d] r6q/bpp2pkn/n2p4/3Pp3/1PP1P3/3B2N1/2Q4p/1R2B2K b - - 0 34

Game 42 a Benko gambit SF let black win back the pawn on b2, But in the meantime SF opened up the kingside. It wins by saccing his rook on h6 on move 26 beautifully (kingside attack win)
[d] 1r3q2/2nnp1bk/r2p2pp/2pP4/P3P3/2NQ1NP1/3B1PK1/R6R w - - 0 26

Game 50 in a Scandinavian mainline white has some advantage because of pressure on the kingside
[d] 5rk1/pp3pp1/2rbpnp1/6N1/2P5/qPB2QP1/P3RP1P/R5K1 b - - 0 24
Komodo ingnores this and playes on queenside with 24..a5. Better was to play Qc5 h4 and Qf5 to try to exchange the queens. After 25 Rd1 a4 26 Bb2 Qb4 27 g4 the game went smoothly into a win for white (a both positionally and kingside attacking win)

Game 58 after a Caro Kann advanced a closed position with white space advantage and an extra double a-pawn and one piece off the board, SF gradually, technically converts this advantage into a win (a kingside attack win)
[d] 5rk1/pp3pp1/2rbpnp1/6N1/2P5/qPB2QP1/P3RP1P/R5K1 b - - 0 24

Game 62 a classical Caro Kann 4..Bf5 mainline SF gets an extra pawn which Komodo for very long time evals as equal because of counterplay. But in endgame the advantage becomes susbtantial enough for a win. (So an endgame win or a technical win)

Game 74 a Richter Rauzer where white cuts like knife through butter with all pieces on board
(a typical quick Sicilian attacking kill)
Allready +1 after move 15 [d] r1b1k2r/2qnbpp1/p2pp3/1p2n1B1/3NP2P/P1N2P2/1PPQ4/2KR1B1R b kq - 0 15

Game 122 a Queens Indian where queens have come off results in difficult position for black. SF managed to win after Komodo decides here to exchange a pair of rooks
(good transversing form middlegame into endgame with positionnaly outplaying)
[d] 2br1rk1/p5pp/1pP1pn2/1P3p2/P1P5/2N3P1/5pBP/R2R3K b - - 0 23

Game 132 SF sees itselfs allready 0.76 out of the book in a classical mainline Kings Indian
After forcing an exchange of queens on move 21 white has +0,85
(positionnaly outplaying)
[d] r1b2r1k/pp4bp/1q3n2/1NQ1p1p1/4Pp2/3N1P2/PP2B1PP/2R2RK1 b - - 0 21

(With black SF holds the game because Komodo tries to force things on the queenside with exchange of white bishop formknight on a6, giving black opportunity for the well known King Indian counterplay by kingside attack)
[d] 1rbq1bk1/p1p3rp/p2p1nn1/2PPp1p1/4Pp2/2NN1P2/1P2B1PP/2RQB1RK b - - 0 20

---------------------------------------------------------------------------
The six Komodo wins

Game 1 a Pirc defence where Komodo went for a strangling win. After 25 f4 +1
(positionnaly outplaying)
[d] 2r1rnk1/1p1qpp1p/2p1b1pb/p7/P2P4/2PBRP1P/1P1N2PB/3RQ2K w - - 0 25

Game 32 another very subtile piece of positionally play or technique in an English game that started rather equally (positionnaly outplaying)
[d] 5r2/1bp3rk/1p1pq1p1/7p/1PP1pP1P/4N1P1/4P1QK/3R1R2 w - - 0 41
Game 55 another Pirc where SF desperately attacks the king (see diagram) but Komodo holds and wins (win by positionnaly outplaying with good defense)
[d] r4r1k/pp5p/n2pp1p1/2p1bP2/4P2q/2NBBPn1/PPP1Q2P/5RRK w - - 0 17

Game 71 a Sicilian Taimanov with white advantage open c-line tripling and better pawns, Komodo after very long manoevring in the middlegame it finally forced a win with an exchange sac 77 Rc7
(positionally outmanoevring)
[d] 1r1b2k1/1q1b1pr1/4p1p1/1p1pP3/1P3P1p/P1R4P/2R1BBPK/2Q5 w - - 0 77

Game 85 a 3Nc3 Bb4 mainline French was for long time rather equal
Than on move 54 Komodo breaks through very nice on the kingside with g4 and gets the upperhand
(kingside attack and pawn passer threats combined)
[d] 2r2r2/q5pk/1pbP3p/2n2p2/p1B1pP2/B3Q2P/2PR2PK/R7 w - - 0 54

Game 125 a Nimzo Rubinstein allready after move 19 black is rather helplessly passive in watching white going for a kill with bishoppair and space advantage and better pieces. With Qh3 alle becomes clear (a kingside attack win)
[d] 4rr1k/pbq2pp1/1pn1p2p/2p1P3/4PP2/P1P3Q1/R5PP/2B1RBK1 w - - 0 20

(to Frank Q, in seven of the nine white won games SF had his queens on the board in the winning position)

Jouni
Posts: 2334
Joined: Wed Mar 08, 2006 7:15 pm

Re: Surprise result 16 cores from CSS forum

Post by Jouni » Tue Dec 29, 2015 10:16 pm

Maybe all SMP patches are finally bearing fruit? Look at framework 15 cores test SF7beta - SF6:

ELO: 144.08 +-6.4 (95%) LOS: 100.0% !! :D
Jouni

Hai
Posts: 548
Joined: Sun Aug 04, 2013 11:19 am

Re: Surprise result 16 cores from CSS forum

Post by Hai » Wed Dec 30, 2015 7:26 am

Jouni wrote:Maybe all SMP patches are finally bearing fruit? Look at framework 15 cores test SF7beta - SF6:

ELO: 144.08 +-6.4 (95%) LOS: 100.0% !! :D
But tested without opening phase.
With opening phase should be +20-40 elo more.

Post Reply