Ninety-seventh KGS Computer Go Tournament

Sunday November 11th, 2013

These results also appear on an official KGS page.


format12-round Swiss
board size19×19
time14 minutes plus 10/30s


The first round started at 08:00 UTC.

Result table

Crazy AyaMC nomiB NiceG Orego mpfue
B12R W16R B110R W11R B17R W111R W14R B19R B15R W18R B13F W112T 126464Winner
2AyaMC W02R B06R W010R
B13R W112R B11R W18R W14R B17R W111R W15T B19T 96933
3nomiBot B01R W07R B011R W03R B012R
W15R B16R W110R B12R W19R B14T W18T 77521
4NiceGo19N B04R W09R W01R B08R B05R W06R B010R
B13R W112R W12 B17 W11122½ 5696
5Orego12 W05R B08R B04R W07R B011R W02R B09R W03R B012R
W1167½ B16T W110T 3750
6mpfuego W03F B012T B05T W09T W04T B08T B02 W07 B01122½ B0167½ W06T B010T
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 16 games and White won 20.

We welcomed a new release of Fuego to this event, playing as 'mpfuego'.


In round 1, mpfuego lost by a large margin to Orego12. Even an old version of Fuego would have expected to beat Orego, so there was clearly something wrong. Mpfuego lost all its games in the tournament, because it was achieving many fewer playouts than it should have been. Its operator later explained "We are using the Message Passing Interface (MPI) library. If an MPI process stops without cleaning up its messages, it will cause troubles for the following processes. I think that is what happened. Normally that should not happen, but it may have happened with the settings we used for the tournament."

NiceGo19N vs mpfuego
At end of "clean-up"

In round 2, mpfuego lost to NiceGo19N. After the players both passed, they disgreed about the status of some stones, and entered the "clean-up phase". After NiceGo19N had played three stones and mpfuego had passed once and played one stone, they each passed again, ending the clean-up phase in the position shown to the right. This was scored as a win to NiceGo19N; this would have been the result if they had both finished the clean-up phase correctly.
      My guess is that mpfuego does not handle the game end correctly, and NiceGo19N was happy to have the game end in a position it knew would be counted as a win for it. But I urge the programmers of both to make them handle the clean-up phase correctly. I have seen many won games lost by bots which failed to do the clean-up correctly.

Orego12 vs NiceGo19N
Moves 1-15

In round 3 the game between Orego12 and NiceGo19N started as shown to the left. This looked odd to me, games betweeen dan-rated players do not start like this. Orego12 then tried to save the stones it had lost in a ladder.

Also in round 3, mpfuego's game is listed as a "loss by forfeit". This happened because its operator logged in to its account to resign for it, causing the server to score it as a loss by forfeit.

In round 4 mpfuego lost on time (in a lost position). It left itself with two seconds for its final move in the Canadian overtime period, and then failed to make this final move in time. It repeated this behaviour in several subsequent games.

The round 7 game between mpfuego and NiceGo19N ended in a similar way to their round 3 game. Both passed when the game was clearly not over, and played only a few moves in the clean-up phase before they both passed again with many regions of the board undecided. Again the game was scored as a win to NiceGo19N, which would also have won if the game had been finished properly.

Annual points

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

Crazy Stone8

Details of processor numbers, power, etc.

Aya, MC version, running on 2 machines: 980X 3.3GHz 6 cores + W3680 3.3GHz 6 cores
Crazy Stone, running on a 24-core server
MP-Fuego, running on 512 cores of a 1024 core Intel Xeon CPU E7-8837 @ 2.67GHz.
oakfoam, running on a mini cluster: i7-2600K + i7-920
nomiBot, running on one machine with 2 Xeon E5-2680 (total of 16 cores, @ 2.70Ghz with hyperthreading).
Orego, running on five nodes of a custom Linux cluster Fido. The node has two AMD Six Core Opteron 2427 2.2 GHz processors (60 cores total), 8 GB RAM, Centos Linux.