Large board go (21x21 and up)

At some point about ten years ago, our national go association purchased sets of go stones that each came with a “paper goban”. They also purchased proper go boards, so the paper boards that came with the stones were pretty useless.

So it was easy to get four 19x19 paper boards, and tape them together to get one 37x37 paper board.

We played a few games. 37x37 is huge and it was obvious that we had no clue about whole-board strategy. Everything felt small and local. I can build a 10x10 or even 16x16 moyo somewhere on the board. My opponent will not invade, but instead build their own moyo somewhere else. Invading their moyo seems too small, consolidating my moyo into territory seems too small, so I’ll build my second moyo elsewhere. And so on. The game takes forever, and during all the “opening”, there is barely any confrontation, because tenuki always seems more profitable than answering locally. There is so much space everywhere, why would I play in a region where there are already stones?


Good questions and observations. I’m playing a few 37x37 games on the Little Golem Game Server (see my previous/ above replies) but have only completely finished one, so I can not answer your questions with any confidence or authority, but you asked:

Maybe it is a bigger move to play elsewhere, but MAYBE attacking the opponent’s moyo is bigger than building up a moyo of your own, AND this would be likely if a move could BOTH expand one’s own moyo and reduce the opponent’s moyo.

I would definitely be interested in trying a unranked corr game on a board larger then the usual board sizes. Say 25x25 larger then that seems too daunting.

Using magic, I have created this magnificent beast of a 49x49 game, as proof of concept:

Game information says that it is ranked, somehow. So, it is certainly possible for the server to create the game.

However, there is a catch: You cannot place a stone anywhere beyond column 25. It would seem that the inner workings of the server are intrinsically tied with the coordinate system. It has no problem with rows because those are just numbers, but it does have problems with columns because they seem to be hardcoded from A to Z (minus the “i”).

This is the error message:

Anyway, it is certainly a shame that it doesn’t work (even though it wasn’t supposed to). I’d love to do some shenanigans with this. Any ideas?


In the series 9x9, 13x13, 19x19, each has roughly twice as many intersections as the last. The next size up is 27x27. Here is a Japanese woodcut print of an actual 27x27 board, from around the 19th century. Note that the star points are on the 5th lines.


Yes, check out the source code, and improve it. Or put in a fix request. It seems to be written in Node.js. I would go with double letters after Z (Y, Z, AA, AB, AC, …) but you could get up to 50 by using single lower-case letters after Z (Y, Z, a, b, c, …) The letter coordinates skip over “I” as too similar to “1”. Or you could use the style shown on yebellz 49x49 board above: A up to Z, followed by 24 down to 1, in both the X and Y dimensions.

1 Like

Wow! I just can’t imagine the pain to carve that grid!

I wouldn’t be so sure.
Prints aren’t photographs. That could be just a concept design. And I actually count 26 intersections on one direction.

But one is on the sixth.
Pointing that out to the author must have been a real pain!


here Game programs for boards more than 19x19 • Life In 19x19 a lot of information about game programs for boards up to 100x100. I have been interested in this for a long time, if you have any questions, write


37x37, b60-s728 - b18-s570 0:1 (both: 1000 visits, chinese, resignThreshold = -0.99)
b60s728-b18s570.sgf (7.9 KB)


Some strange moves at the very beginning. … like first line ??

1 Like

networks have not been trained on such large boards, so there are unusual or strange moves. probably the network is “disoriented”. but interestingly, the 20b and 18b networks feel comfortable on large boards

in the second game 18b plays black and another beginning
37x37, b18-s570 - b60-s728 0:1 (both: 1000 visits, chinese, resignThreshold = -0.99)
b18s570-b60s728.sgf (7.9 KB)


37x37, b18-s770 - b60-s897 (b18 100 visits, b60 50 visits, chinese, resignThreshold = -0.999)
katago1_Vs_katago2.sgf (7.4 KB)

18b for some reason incorrectly estimates the account
note: I changed my username


Thats awesome! Most of the opening moves feel really slow on that gigantic board, but i guess most of those were either sentes or reverse sentes. Interesting game ^^


It’s not only long to play but long to watch too😂
What i noticed is: extensions are more something used by AI as in a 19x19.
There is another recurring process where a color build a wall in exchange for points, but then the other color cut directly into the weakness of the wall and so on…
And tenukis seems to be less numerous.
Because of the size i was expecting some funny things toward controlling the center but i didn’t catch anything special, i am too weak to understand that side maybe?

1 Like

Besides experimentation with AIs i am curious to how long it would take for two humans to play a game like one you play without clock on a 19.
A day, a week, a month, more?


I made a 31x31 board and have played four games on it with a few people. I acquired three sets of old slate and shell (from badukclub) to play on it.

I chose 31x31 because the extra star points can be spaced apart at the normal distances (6 lines), while still having them on the 4-4s and tengen, and the next largest size by those parameters is too large to reach across while seated.

I was inspired to make it by an article about a 43x43 board in the October 2021 issue of the European Go Journal.

Playing at a fairly rapid pace, I think the games have taken an hour or two, but none have gone to counting yet.


I played a 5th game on the 31x31. As 4 kyu versus 7 kyu, I gave a 9-stone free placement handicap, with which he made shimaris in four corners, and black won by just 2.5 points! That was the first one I’ve played to go to counting.