Ninety-fourth KGS Computer Go Tournament

Sunday July 7th, 2013

These results also appear on an official KGS page.


format7-round Swiss
board size19×19
time29 minutes plus 10/30s


The first round started at 16:00 UTC.

Result table

Zen19 AyaMC ManyF gnugo gomor nomiB NiceG Orego
W11R B17R W12R B15R W13R W14R W16R 72626Winner
2AyaMC B01R W07R
B14R W12R W15R W13R B16R 52713
3ManyFaces1 B02R W04R
W1132½ B1729½ W16R B13R W15R 52412
4gnugo3pt8 W05R B02R B0132½ W0729½
W16 W14R W13R 3275
5gomorra3 B03R B05R B06R
B11R W07R W12R B14R 3255
6nomiBot B04R W03R B06 W01R B17R
B15R W12R 3235
7NiceGo19N B06R B03R B04R B02R W05R
W11R B17R 2210
8Orego12 W06R B05R B03R W04R B02R B01R W07R

Excluding forfeited games, Black won 11 games, White won 17 games, and there were 0 jigoes.

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 19 games and White won 17.

Seven players registered for the tournament. I therefore added GNU Go, running on a single processor of my own desktop PC. I believe that having this play is better than having byes. I believed that GNU Go was so weak that it would be unlikely to affect the result, but strong enough that it can win the occasional game. However, see below.


I have nothing to say about the actual play, as most of the players are stronger than me.

In round 4, gnugo3pt8 (rated at 6-kyu on KGS) beat nomiBot (rated at 2-dan). This surprised me. The game seemed unremarkable.

gnugo3pt8 vs NiceGo19N
After move 209

In round 6, gnugo3pt8 beat NiceGo19N (rated at 1-kyu). This surprised me more. Again the game seemed unremarkable, with gnugo3pt8 making some poor moves, as I would expect from a 6-kyu player. NiceGo19N's operator admitted that it sometimes loses to version 3.9 of GNU Go when he is testing it.
        I wonder if the reason that nomiBot and NiceGo19N lost to a much weaker player, is that they are "trained" mostly against other Monte-Carlo bots, and have little experience of playing against old-style bots like GNU Go. But as the ratings of all three are derived from their games against human opponents, this is not a convincing explanation.

Oakfoam (playing as NiceGo19N)'s programmer Detlef Schmicker has responded

It was one of the observers who tested oakfoam against gnugo 3.9, not me. Our actual download version is missing the large pattern database, so it is not comparable I think.

We read the top left in the game not dead enough (90% dead). as it is a big area this lead to significant miscount. As we do not have dynamic komi oakfoam gave away a lot of points, as it had 80% winrate for most of the middle game.

I think the probability of such problems is not so dependent on the opponent strength. If playing against even strong humans this happens from time to time and loosing does not harm the rating too much I think.

By the way, I am training mainly against gnugo level 0 and gnugo level 10. But never with mc enabled.

The diagram to the right shows position after Black played tenuki from the fight in the top left corner. It appears to me that the black group in the top left has no way to make a second eye, and that the game is very close.

Annual points

Players receive points for the 2013 Annual KGS Bot Championship as follows:

Many Faces of Go4

Details of processor numbers, power, etc.

Aya, MC version, running on 980X 3.3GHz 6 cores
GNU Go version 3.8, running at its "level 16" on a single 3.3GHz Intel i5-2500 CPU.
Gomorra, running on an OCuLUS cluster, using presumably 40x16cores. Cores are running at 2.6 Ghz
Many Faces of Go, running on an Amazon EC2 Cluster Compute Eight Extra Large Instance, 2 x Intel Xeon E5-2670, using 16 cores, 16 threads, and 48 GB of memory.
oakfoam, running on an i7-2600K
nomitan, running on 2 Xeon E5-2680 (total of 16 cores @ 2.70Ghz with hyperthreading).
Orego, running one of the five nodes of a custom Linux cluster. The node has two AMD Six Core Dual Opteron 2427 2.2 GHz processors (12 cores total), 8 GB RAM, Centos Linux.
Zen, running on a mini-cluster of a quad 16-core Opteron 6376@2.3 GHz, two dual 6-core Xeon X5680@4 GHz, a 6-core i7 3930K@4 GHz, a 6-core Xeon W3680@4 GHz, and a 6-core i7 980X@4 GHz computers connected via a GbE LAN. 6 computers, 106 cores total.

13-06-02 12:17 gghideki Room 354 The opening was, I guess, good for Aya. 13-06-02 12:21 gghideki Room 354 E7 (18th move) looks overplay RayTomes [1d]: g2 wins