A Played Game

In the game below, it was not me, who played against GambitVB, it was a special game-computer. I just entered the moves. I choose this game, because GambitVB played straightforward and more or less according my intentions. It does not go always that way; not yet.

ECO: B27
Opening: Sicilian defense
Variation: Hungarian

    W.Rens  GambitVB
  1 Ng1-f3     g7-g6  00:00:21 00:00:00
  2  e2-e4    Bf8-g7  00:00:30 00:00:00
  3  d2-d4     c7-c5  00:00:39 00:00:00
  4 Bf1-c4            00:02:21

These first three black moves came out GambitVB's openings book. In this book, it finds pre-programmed opening variations with the corresponding information that names the variation. The target length of the variations in that book is six moves for GambitVB; sometimes it is less, sometimes more.

GambitVB played 4 . . . c5xd4 at this position, for which it used 48 seconds. In that timespan, it looked up to ten moves deep, for which it generated 261,000,000 moves; that is 5,400,000 moves per second.  GambitVB used those moves to build a virtual movetree. Virtual, because it is not realistic to remember and reuse all those moves. GambitVB looks just for variations according a rigid algorithm and remembers the best variation. It does store a sub tree of the virtual move tree, only for moves it expects to be reusable and promising according its current knowledge. In this case, for 4 . . . c5xd4, it created 5,200,000 tree node objects, that is 2% of the 261,000,000 moves it generated.

In addition, in the same timespan of 48 seconds, GambitVB generated 273,000,000 moves only for captures and pawn-promotions, used for material analyses. GambitVB executed also 379,000 strategy analyses for the best looking moves; only, 0.15% was selected from the generated moves. This strategy analysis looks for things like the strength of the pawn formation and the safety of the king.

  4            c5xd4           00:00:48
  5 Nf3xd4    Qd8-c7  00:03:03 00:01:34
  6 Qd1-d3    Bg7xd4  00:03:52 00:02:23
  7 Qd3xd4    Ng8-f6  00:04:18 00:03:12
  8 Bc1-e3    Nb8-c6  00:05:03 00:04:00
  9 Qd4-c3       0-0  00:05:12 00:04:47
 10 Be3-h6    Rf8-e8  00:06:09 00:05:38
 11 Nb1-d2    Qc7-e5  00:07:07 00:06:30
 12 Qc3xe5    Nc6xe5  00:07:22 00:06:33
 13 Bc4-b3     b7-b5  00:07:44 00:07:23
 14 0-0       Bc8-b7  00:08:50 00:08:18
 15 a2-a4     Ne5-g4  00:09:05 00:09:11
 16 Bh6-g5     b5xa4  00:09:17 00:10:05
 17 Bb3xa4     a7-a5  00:10:48 00:11:01
 18 Bg5xf6     e7xf6  00:12:33 00:11:03
 19 Ba4xd7    Re8-d8  00:13:50 00:11:06
 20 Bd7xg4    Rd8xd2  00:14:25 00:11:08
 21 Rf1-c1    Bb7xe4  00:15:59 00:11:09
 22 Bg4-d1            00:17:09

GambitVB played 22 . . . a8e8, for which it used 66 seconds. In that timespan, it generated 915,000,000 moves, which is the highest amount per move in this game. In the same time, it used a record amount of tree nodes in memory: 40,000,000. To store those tree nodes, it used 1.608,000,000 bytes of memory. That is 52% of the memory available for usage at that moment.

 
 22           Ra8-e8           00:12:17
 23  h2-h3    Be4-d5  00:18:49 00:13:24
 24  f2-f3    Re8-e1+ 00:19:13 00:13:27
 25 Kg1-h2    Bd5-c4  00:20:40 00:13:31
 26  h3-h4    Bc4-f1  00:22:22 00:13:38
 27  f3-f4    Bf1xg2  00:23:53 00:13:40
 28  b2-b4    Re1-h1+ 00:25:20 00:13:41
 29 Kh2-g3    Rh1-h3+ 00:25:29 00:13:41
 30 Kg3-g4     h7-h5+ 00:25:41 00:13:41

GambitVB foresaw a pawn gain for the moves 24...Re1+ (3 seconds thinking time), 25...Bc4 (4 seconds), and 26...Bf1 (5 seconds). It foresaw a pawn plus a bishop for move 27...Bg2 (3 seconds of thinking time) and mate from move 28.