Hundred and sixteenth KGS Computer Go Tournament

Sunday October 4th, 2015

Rules

Times

The first round started at 16:00 UTC.

Result table

Black won 38 games, White won 38, and there were 4 jigoes.

Players

Eight players registered. We welcomed a new player to these events, CGI, playing as CGI9. "CGI Go Intelligence" is a Taiwanese program, with multiple developers, and ancestors in Amigo and HappyGo: See its history.

Results

I had expected that, with the entrants being much stronger than me, particularly at 9×9 Go, I would have little to say about the games. In fact, in round 1, all four games gave me something to write about.

fuego9 vs abakus
After the last move of the game.

• CGI9 did not move in its game with CrazyStone. After five minutes, its operator asked me for permission to re-start it, which I granted. It then started to play normally, but with most of its time already gone, it lost. Apparently it had failed to receive from the server the message telling it to start playing.
• AyaMC, in this round and throughout the tournament, disconnected for several minutes at the start of the game and again after each stone it played. It therefore played about two stones in each game, and lost all its games on time. Its creator and operator Hiroshi Yamashita later explained that this happened as a result of his using kgsGtp 3.5.20. When, after the tournament, he switched back to using kgsGtp 3.5.10, it played normally.
• When abakus played the move marked in the diagram to the right, its opponent fuego9 resigned. Apparently it realised that its right-hand group is dead by "bent-four-in-the-corner".
• MCark and NiceGo19N ended their game by passing consecutively, then disagreed about the status of some stones, so the game entered the clean-up phase. NiceGo19N did not make any movers in the clean-up phase, and eventually resigned. In many of its subsequent games, NiceGo19N behaved a bit differently: when it thought it had lost it did not resign, because of a newly-introduced bug in its code, but passed. At the game end it disagreed about the status of some stones, and in the clean-up phase it tried to resign, but its resignation was rejected by the KGS server, and it eventually lost on time.

abakus vs CrazyStone
Move 58.

In its round 3 game with CrazyStone, abakus started a ko fight with the move shown to the left. Both players fought the ko competently, and abakus eventually won the game.

In round 12 abakus lost its first game, to CrazyStone.

In round 13 abakus lost its second game, to fuego9.

fuego9 vs NiceGo19N
After consecutive passes.

In round 16 fuego9 and NiceGo19N both passed in the position shown to the right, and agreed to mark the dead stones as shown, leading to a win for fuego9. My guess is that NiceGo19N passed because it knew it had lost (but has forgotten how to resign, as mentioned above), while fuego9 understood that passing in the position shown would give it a win. My congratulations to both, for achieving a correct result without a slow cleanup!

CrazyStone vs abakus
Move 39.

In round 19, abakus played the move shown to the left, against CrazyStone. This seems to me to achieve nothing, its only merit is that it keeps sente. But I suspect that Black has no way to win, so this move is as good as any.

Annual points

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

Details of processor numbers, power, etc.

abakus

abakus, running on 5 nodes each with two Intel Xeon E5-2670, 2.6GHz and 64 GByte main memory (80 cores total).

AyaMC

Aya, MC version, running on an i7 980X 3.3GHz, 6 cores.

CGI9

CGI Go Intelligence running on "8 x 3 x AMD Opteron 6174/2.2 GHz (ALPS system by Acer)", each one has 36 cores, total 288 cores.

CrazyStone

Crazy Stone, running on Amazon ec2 c4.8xlarge server instance (xeon, 18 cores, 36 threads, 2.9GHz).

fuego9

Fuego, running on 12x Intel(R) Xeon(R) CPU X5670 @ 2.93GHz.

MCark

MC_ark, running on a Xeon E5-2450 2.10GHz 8-core HT on*2cpu + GPU(G200eR2).

NiceGo19N

oakfoam, running on an i7-4790K / GTX-970.

Orego32

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.