Hundred and twelfth KGS Computer Go Tournament

Sunday May 3rd, 2015

These results also appear on an official KGS page.


format12-round Swiss
board size19×19
time9 minutes plus 10/30s


The first round started at 16:00 UTC.

Result table

Zen19 DolBa AyaMC NiceG ManyF gnugo
W13T B16T W110R B12R W17R B111R W14R B18R B15R W19R B11R W11220½ 126464Winner
2DolBaram B03T W06T B010R
W14R B112R W15R B17R W111R B11R W18R B12R W19R 97034
3AyaMC W02R B07R W011R B04R W012R
B11R W19R W13 B16R W110 W15R B18R 77420
4NiceGo19N B04R W08R B05R W07R B011R W01R B09R
B02R W112T W13 B16 W110 4734
5ManyFaces1 W05R B09R W01R B08R B03 W06R B010 W12R B012T
W1420½ B1731½ W11124½ 4714
6gnugo3pt8 W01R B01220½ W02R B09R B05R W08R B03 W06 B010 B0420½ W0731½ B01124½

Black won 15 games and White won 21.


Five players entered. I therefore added GNU Go, to make the numbers even.


In round 3, DolBaram suffered from lag and disconnects in its game with Zen19X, and lost on time. Its operator then moved it from its hardware to run on the Amazon cloud.

After round 5, Zen19X had won 5 games, DolBaram 4, AyaMC 3, ManyFaces1 2, NiceGo19N 1, and gnugo3pt8 0.

In round 6, DolBaram reached a very close endgame in its game with Zen19X, and lost on time while playing the final stone of its first "10 stones in 30 seconds" overtime period.

After round 11, Zen19X had won 11 games, DolBaram 8, AyaMC 7, ManyFaces1 4, NiceGo19N 3, and gnugo3pt8 0, each player having won all its games against lower-scoring players. This list of scores puts the second player's score close to that of the third, and the second-to-last's close to that of the third-to-last. This closeness does not reflect their relative strengths, but the way in which the KGS pairing algorithm makes the second player play more games against the first player than against the third (and the second-to-last more against the last than against the thrid-to-last). I believe the pairing algorithm is correct to do this in a Swiss tournament; but I caution tournament organisers about the use of the Swiss format, if the allocation of second prize is to be important.

In round 12, ManyFaces1 stopped playing when it had more than 10 of its original 15 minutes left on its clock, and lost on time to NiceGo19N, changing the final order.


MC-based Go-playing programs are perceived by humans as playing "high", as if trying to make central territory. This is particularly true of Zen, which is perceived as playing high even in its games with other MC-based programs.

However, look at these three diagrams below, all of games between Zen19X and DolBaram.

Zen19X vs DolBaram, after move 32 DolBaram vs Zen19X, after move 38 Zen19X vs DolBaram, after move 29

In the first and third figures, Black appears to be making a central moyo, and in the second White does. In each case, that player is DolBaram, and it is Zen who has more corners. So it is now DolBaram that deserves the reputation of using Cosmic Style.

Annual points

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

Many Faces of Go1

Details of processor numbers, power, etc.

Aya, MC version, running on an 980X 3.3GHz, 6 cores.
DolBaram, running on a dual Xeon-X5660, 12 cores at 2.8GHz for rounds 1 to 3, and Amazon C3.8xLarge (2.8GHz * 16core) for rounds 4 to 12.
GNU Go, version 3.8, running one thread on one i5-5200 CPU.
Many faces of Go, running on "an old local machine", as its operator did not have time to set it up on the Amazon cloud.
oakfoam, running on an i7-4790K / GTX-970.
Zen, running on a dual 10-core Xeon E5-2690 v2@3 GHz 32 GB RAM.