Hundredth KGS Computer Go Tournament

Sunday February 2nd, 2014

These results also appear on an official KGS page.


format21-round Swiss
board size9×9
time9 minutes plus 10/30s


The first round started at 16:00 UTC.

Result table

Crazy Zen19 DolBa pachi AyaMC NiceG Cyrag MCark
W11T B172 W1122 B018R B122 W196 B1132 W1192 W13R B111R W117R J21 B15R W110R B115R W18R B114R W14R B120R W16R B116R 19½219½194Winner
2Zen19S B01T W072 B0122 W118R
J4 W1102 J14 J16 B0201 W15R B19R W12R B0111 W113R W16R B115R W121R B13R W117R W18R B119R 14½237½111
3DolBaram W022 B096 W0132 B0192 J4 B0102 J14 J16 W1201
W16R B18R W115R W17R B121R B15R W117R B118R B11R W111R W13T B112R 14½237½102½
4pachi B03R W011R B017R J21 B05R W09R B06R W08R B015R
B11R J14 B019R W120R B14R W110R B17R W112R B116R B12R W113R B118R 11227½57
5AyaMC W05R B010R W015R B02R W1111 B013R B07R W021R W01R J14 W119R B020R
W13R B112R W016R B16R W18R B118R B14T W19R B117R 10½21761½
6NiceGo19N B08R W014R B06R W015R B021R W05R B017R W018R W04R B010R B03R W012R B116R
W02R B19R W113R B019R W01R B17T W111R B120T 6211½30½
7Cyrago9 B04R W020R W03R B017R W01R B011R W07R B012R W016R W06R B08R W018R B12R W09R B013R W119R
W15T B010R W014R B015R W121F 4205½20
8MCark B06R W016R B08R W019R B03T W012R W02R B013R W018R W04T B09R W017R B11R W07T B011R W020T B05T W110R B114R W115R B021F
In the table above,
   0 is a loss
   1 is a win
   J is jigo
   left superscript is the player's colour
   right superscript is the round in which the game was played
   a subscript shows how the result was determined:
      R for resignation
      T for time
      F for forfeit
      a number for the points difference after counting.
All the 0s, 1s and Js are links to the game record.

Black won 37 games, White won 41 games, and there were 5 jigoes.


I was very pleased with this, the 100th KGS bot tournament. (In fact the numbering is not really correct, this was the 100th monthly KGS bot tournament excluding the "Slow" tournaments.) The entrants included what appear to be the world's three strongest programs, and I watched the games with interest.

I have often said in recent reports that, as a 3-kyu player, I do not feel able to comment on games by players which have dan ratings. This ought to be particularly true of 9×9 Go, at which the strongest programs are said to be near professional strength. Yet I noticed several moves by strong players in this tournament which seemed to me to be blunders. I have described them below.

In round 1, CrazyStone played Zen19S. After all meaningful moves had been made, CrazyStone was ahead by two points. But it continued making meaningless moves, while Zen19S passed. However one of Zen19S's passes was not received by the server, and Zen19S is shown as losing on time.

In round 2, Zen19S had hardware trouble with one of the three miscellaneous computers that it runs on, and was several minutes late joining its game with AyaMC. It still won.

DolBaram vs CrazyStone
Moves 37-41

Also in round 2, before the sequence shown to the right was played, DolBaram as White had a winning position against CrazyStone. However, ajahuang has pointed out that move 34 is a blunder. Move 38 is also a blunder; instead White can play 38 to the north of 35, allowing him to recapture one stone when Black captures four and keeping his group connected.

In round 3, MCark vanished after playing one move against DolBaram, and lost on time. It continued to have problems up to round 7, disconnecting and losing on time in rounds 3, 4, 5 and 7.

CrazyStone vs pachi
Moves 51 and 52

Also in round 2, pachi blundered against CrazyStone, as shown in the figure to the left. If, instead of 51, it had connected its dead stones in the bottom left making a "bulky five" shape, CrazyStone's left side group would have been reduced to one eye, giving pachi a clear win.

Zen19S vs pachi
At end of game

In round 5 the game betwen Zen19S (White) and pachi ended in the position shown to the right. Zen19S has won by 88 points.

Cyrago9 vs pachi
Move 41 Moves 63-65 Move 85

In round 7, pachi, in its game with Cyrago9, entertained the observers with some of its moves, as shown in the three diagams above. With move 41 (left diagram) it failed to make a "bulky five" at the top left, which would have killed the white group. This was the same mistake that lost pachi its game against CrazyStone, shown above. But Cyrago9 failed to take advantage of this and save its group.
       With move 63 (middle diagram) it again failed to make a "bulky five", this time in the bottom right, which would eventually have killed the other white group. Maybe it judged that making the connection in the centre with 63 was more urgent, as, with two kos, the lower right white group cannot be captured quickly. Again Cyrago9 failed to save its group. With move 65, pachi surprised the observers by correctly making the "bulky five".
       With move 85 (right diagram) finally made a "bulky five" in the top left as well, leaving all the white stones dead.

At the end of round 7, MCark's operator successfully updated his version of the kgsGtp.jar client. MCark ran correctly for the next twelve rounds, without disconnecting. I was puzzled by this. I had been running a copy of GNU Go correctly before the tournament (to test it in case it was needed to make the numbers even), and did not encounter any problems with the client. I was running version 3.5.0, dated 2011-07-24.

In round 14 there were two jigoes, between DolBaram and Zen19S, and between pachi and AyaMC.

Zen19S vs CrazyStone
After last meaningful move. Moves 75-77

In round 18, CrazyStone lost to Zen19S, having won its previous 17 games. This was another interesting game. CrazyStone may have thought that it had killed the white group in the top left, and that it was winning when the last "real" move was made, as shown in the left-hand diagram. But the white group is alive in seki, and White is ahead.
      Instead of passing, CrazyStone then played purposeless moves inside Zen19S's territory, while Zen19S passed. Eventually, Zen19S passed when it should have responded, and CrazyStone was able to start a ko with move 75, as shown in the right-hand diagram. If it had then fought this ko, by playing on the top right corner point and playing self-atari in the seki as a ko threat, it could have won the ko, and (I believe) won the game by two points. Instead it conected the ko with move 77, and lost the game.

DolBaram vs ZeInky n19S
At end of game

In round 20, Dolbaram (White) and Zen19S both passed in the position shown to the left. DolBaram thereby won by one point – an odd margin if victory is possible because there are three neutral points in the seki at the bottom left. If Zen (Black) had played at the point marked with a triangle, it would have claimed one of these points, making the result jigo. (White cannot claim any of the three points without losing his whole group).
      Towards the end of round 20, MCark disconnected, and lost on time, again. It also lost, by Forfeit, in round 21.

Annual points

Players receive points for the 2014 Annual KGS Bot Championship as follows:

Crazy Stone8

Details of processor numbers, power, etc.

Aya, MC version, running on two machines: 980X 3.3GHz 6 cores, and W3680 3.3GHz 6 cores.
Crazy Stone, running on a 24-core server
Cyrago, running on a single thread of a Q9550 @ 2.8 GHz.
DolBaram, running on a Xeon E3-1230 v3 (3.30GHz 4core).
MC_ark, running on Xeon E5-2687W 3.10GHz 8core HT off *2cpu*2machines + Core i7 870 2.93GHz 4core 8thread.
oakfoam, running on a mini cluster: i7-2600K + i7-920.
pachi, running on one 8-threaded AMD FX(tm)-8350
Zen, running on a mini cluster of a dual 10-core Xeon E5-2690 v2@3 GHz 32 GB RAM, a dual 6-core Xeon X5680@4 GHz 24 GB RAM, and a 6-core i7 3930K@3.2 GHz 16 GB RAM, computers connected via a GbE LAN. 38 cores total.