Autumn 2015 Slow KGS Computer Go Tournament

September 6th-9th, 2015

These results also appear on an official KGS page.


format8-round Swiss
board size19×19
time235 minutes plus 10/60s


The first round started at 22:00 UTC.

Results table

abaku AyaMC Zen19 NiceG Orego gnugo
W13R B07R W12½ B16R B04R W18R B11R W15 63324Winner
2AyaMC B03R W17R
B01R W15R B16R W14R B18R B1217½ 62917
3Zen19S B02½ W06R W11R
W13R B15R W17R W1432½ B1827½ 62715
4NiceGo19N W14R B08R B05R W06R B03R
B12R W01R B17T 3349
5Orego32 W01R B04R W08R W05R B07R W02R
W13½ B16 2352
6gnugo3pt8 B05 W0217½ B0432½ W0827½ B11R W07T B03½ W06
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.

Excluding forfeited games, Black won 11 games, White won 13 games, and there were 0 jigoes.

Five players registered. The organiser added gnugo3pt8 to make the numbers even and avoid byes.


The round 2 game between abakus and Zen19S was interesting. Most observers thought that Zen19S was winning, but overplayed. Abakus won. As both players were probably stronger than any of the observers, and certainly much stronger than me, I will make no comments.

NiceGo19N vs Abakus
Moves 72-97. Move 98 captures at 74.

In the round 4 game between NiceGo19N and abakus, abakus misread a ladder, as shown to the right. Its loss was large, and it lost the game.

Zen19S vs Abakus
Move 136.

In the round 6 game between Zen19S and abakus, Zen19S played a poor move on the 1-line, as shown to the left. This may have a similar explanation to a poor move it made in the previous Slow KGS bot tournament, which was explained as possibly caused by an integer overflow.

AyaMC vs Abakus
Moves 71-84.

In the round 7 game between AyaMC and abakus, abakus played moves 81 and 83 as shown to the right. To even a weak human player, these are clearly bad moves: when White answers with the obvious 82 and 84, White is still winning the semeai, Black has two more dead stones, and White has some useful outside strength. And, I believe, this analysis is correct. This puzzles me. I realise that an MC/UCT program like abakus does not count liberties the way I do, but I have observed that its middle game is usually very strong: it not only "outreads" me, it "outreads" Zen. And I don't see anything in this position, such as seki or multiple ko, that might confuse it.

Also in round 7, gnugo3pt8 lost on time after the computer it was running on suffered an electricity outage.

In round 8, abakus got its revenge on NiceGo19N, with NiceGo19N misreading a ladder this time.

Annual points

Players receive points for the 2015 Annual KGS Bot Championship as follows:


Details of processor numbers, power, etc.

Abakus, running on 8 x 16-core nodes of OCuLUS (
Aya, MC version, running on an i7 980X 3.3GHz, 6 cores.
GNU Go, version 3.8, running one thread on one i5-5200 CPU.
oakfoam, running on i7-4790K.
Orego, running on an instance in the Google Compute Engine cloud (type n1-highcpu-32, which has "32 virtual CPUs and 28.8 GB of memory"). The machine runs Centos 7.
Zen, running on a mini cluster of a dual 12-core Xeon E5-2690 v3@2.6 GHz 32 GB RAM and a dual 10-core Xeon E5-2690 v2@3 GHz 32 GB RAM. Computers connected via a GbE LAN. 44 cores total.