Second KGS Slow Computer Go Tournament

October 11th - 15th

This is a slow tournament. The players each had 12 hours for their moves, and the rounds were 24 hours apart.

These results also appear on an official KGS page which links to the records of all the games.


format5-round Swiss
board size19×19
time12 hours sudden death


The first round started at midnight UTC.

Results table

3rdSimpleBot (Zen)31415
7thPNUGo (valkyria)1130

Ten players registered: AyaMC, CzechBot (which is MoGo), ManyFaces2, Orego12, pachi, PNUGo (which is GNU Go), PueGo (which is Fuego), SimpleBot, valkyria, and Zen19S.

In setting up the tournament, I left enabled the setting "Drop Latecomers". This means that players which are not connected to the server two minutes before play starts are automatically removed from the tournament by the server. This is reasonable for normal tournaments, and I had forgotten its existence. It was not appropriate for this tournament, and unfortunately caused valkyria19 and Zen19S to be dropped. In future, I intend to disable this setting ‒ I can always remove absent programs from the tournament myself if I do not expect them to appear soon.

People were keen to see valkyria19 and Zen19S compete. Aloril agreed to let Zen19S use SimpleBot's account, and this was arranged during the first round, unfortunately after SimpleBot had lost a corner. Petr Baudiš agreed to let Zen19S use PNUGo's account; this was arranged after PNUGo had finished its first-round game. I thank Aloril and Petr for their generous cooperation.


SimpleBot vs CzechBot
Position after move 36.

In round 1 SimpleBot played CzechBot SGF. However as explained above, SimpleBot gave up its place in the tournament to Zen (which had intended to play as Zen19S). When Zen took over, the position was as shown to the right. SimpleBot has lost a huge corner in the lower left, giving Zen a big disadvantage, so it is not surprising that Zen lost this game.

Before the start of round 2, valkyria started to use the PNUGo account. PNUGo had not been expected to do well, as it does not use MC code, and so does not scale well and would not benefit much from the long time limits.

Zen (playing as SimpleBot) and valkyria (playing as PNUGo) had an interesting game SGF with a complicated fight, involving multiple kos, in the lower left corner from moves 76 to 98. Zen came better out of this, and eventually won.

AyaMC vs pachi
Moves 121-126.

In round 3 AyaMc and pachi played the game shown to the left SGF, with a large semeai between the two triangled groups. After Black 121, the white triangled group has four liberties and the black triangled group has three; and it appears that Aya as White thought this was good enough, and tenukied with 122. However the ko in the bottom right corner means that while Black can keep the ko open, White needs an extra move to fill all Black's liberties. It is as though White realised this after 125, for 126 is another tenuki. Aya could not recover after this large loss.

ManyFaces2 vs CzechBot
Position after move 164

      Also in round 3, SGF ManyFaces2 lost a large central group to CzechBot, as seen to the right, but was (in my opinion) still ahead. However some kibitzers believed that CzechBot was ahead at this point, and somehow blundered away its lead. It is possible that they were misled by the "Score Est" feature on KGS, which counts twice the correct value for captured stones.

ManyFaces2 vs SimpleBot (Zen)
Moves 225-227.

In round 4 ManyFaces2, with no loss, played SimpleBot (in fact Zen) SGF whose only loss was the game started by SimpleBot. In the position shown to the left, 225 threatens ten white stones. Zen could have saved them by playing 226 at S13, but if it had, it would have lost the game by a few points. It therefore preferred to play 226 as shown and invite a ko. When Manyfaces2 started the ko with 227, Zen must have realised it was going to lose it, and resigned.

AyaMC vs PNUGo (valkyria)
Moves 205-207.

      Also in round 4, AyaMC played GNUGo (in fact valkyria). SGF. In the position shown to the right, valkyria must save its group at the upper left. It can do this by playing at L19, or, less effectively, at K18. However, it postponed the decision on how to play at the top with a kikashi at A3 (which gives away several points) before playing K18.


Details of processor numbers, power, etc.

Aya, running on 8 cores of a Xeon 5355 2.6GHz
MoGo, running on 4 threads of an i7 920
Many Faces of Go, running on a 12-core Xeon.
Orego, running on one of the five nodes of a custom Linux cluster build by PSSC Labs:
pachi, running on an 8-thread i7 920 box
Fuego, running on two threads on a Core2Duo E7200
GNU Go, unspecified platform
running on one processor of a 4GiB RAM, AMD Athlon(tm) 64 X2 Dual Core Processor 4000+, shared with WeakBot50k
valkyria, running on a i7-860 4 Core processor at 2.80 Ghz.
Zen19S, running on 5 of the 6 cores of a 4 GHz Xeon W3680.