What's the maximum core Zen support? Is it 10 logical cores as I expected earlier?

Leela is a nice program and open source. Kudos to the developer.

But I prefer the Zen GUI for analyzing my games.

I think there is no setting for seconds per move in Leela?

Yes Zen interface is better, that's the only weak point of Leela I guess.

And no unfortunately there are no setting for seconds per move but with the pondering it's not that important because Leela thinks during the turn of the opponent. Also you can change the time setting during the game to restablish it to the initial one. For example if you put 20 minutes for Black and White Leela will play something like 15s a move at the beginning and then it will decrease to 14, 13, etc. So when you'll click on settings and time you'll see it's written less than 20 minutes because the time has passed, so you just have to put it back to 20 minutes if it is written 15 minutes for example to once again having Leela playing 15s a move as at the begninning.

Unfortunately there is no setting in the GUI, but it can be set in the text UI using GTP commands. “time_settings 0 30 1” sets 0 seconds main time, 30 seconds byoyomi time for 1 stone. It’s annoying to play in the text UI, I did it only for a first testing game. It would be far better if Leela would allow to give initial commands as command line arguments, or to have a clock in Sabaki. Or a GUI must be able to send configurable initial GTP commands.

And than it would be nice if a main time could be set in Zenith. But this isn’t possible too. By the way: the maximum thinking time per move is still limited to 120seconds. No custom settings are possible.

Here are some more games.

I cannot remember Zenith 6 won even a single game against Crazy Stone FE on my PC, but Zenith 7 does.
Intel i7-7700K@4.2GHz
NVIDIA GTX 1080

CSFE (30sec/move) – Zenith 7 (30sec/move)


Zenith 7 (30sec/move) – CSFE (30sec/move)


And someone ask for handicap games against CSFE 5d. Here are two games with different settings, this time placed on an Intel i7-6700HQ @ 2.6GHz.

CSFE (5d) – Zenith 7 (30sec/move)


For the second game I set Zenith 7 to 5sec per move. It used about 8 minutes in total. CS with 5d settings used about 1 minute in total. In this game CS evaluated the positions totally wrong, a very long time it didn’t see that whites center group was alive.

CSFE (5d) – Zenith 7 (5sec/move)

Hi Bas,
We have similar hardware, and consistent results.
Very tentatively, with good CPU and gtx1080 it seems that :

Z7 >>> CSDL5d
AQ gpu >>>>> CSDL5d
Leela11 GPU >> Leela11
AQ gpu >> Leela11 GPU
Z7 > Leela11 GPU (but tartaric seems to strongly disagree on this one )

That would be, according to CSDL ranking, something like :

Z7 8d
L11 gpu 7-8d
L11 5-6d
AQ gpu 10d

All these seem very strong, even if CSDL 5d is not really 5d.

I'm looking forward to a lot more games with different hardwares (mmmmh.... q30 maybe, thx )

Is there anyone willing to try out Zen 7 vs AQ? I have Zen but I don't have strong hardware for AQ.

Zenith Go 7 vs Leela 11 GPU (Second Game)
Intel 2500k @ 4,2 Ghz vs Nvidia 1080 Ti
1 : 0
30 sec. p. move vs 20 min. p. game

(Total Score 1:1)



Zenith Go 7 vs Leela 11 GPU (First Game)
Intel 2500k @ 4,2 Ghz vs Nvidia 1080 Ti
0 : 1
20 sec. p. move vs 20 min. p. game

Big KO decides this one. Quite interesting, Zen 7 still thinks it is leading at the start of the Ko fight (Leela 11 is self assured too). But if you let Zen analyze a little longer,

Can you also attach .zgf files? It seems to me that in the first game Zen would've won if it started playing that ko from the other side (m1). Maybe it needed extra 10 seconds
Zen seems to be much stronger on 30 sec or 60 sec and more...

ZEN HAD BEATEN LEELA ON 2 STONES.

Zen killed big Leela's group in a spectacular fashion, although he didn't have to. Zen managed to turn the game around in 160 moves, after which it was very easy.

Zen: Max time per move was 2 minutes. Average spent time per move was 50 seconds. Running mid class Intel i5 CPU.
Leela: "Running 4 (out of 16) threads on a Ryzen 1700 CPU and 1080 Ti GPU" (RoyalLeela on OGS, thanks to roy7).

Attached are clean sgf, Leela's evaluation sgf, and Zen's zgf with evaluations and thinking times.

Edit: Since zgf can't be uploaded, here's a link to zgf (expires after 7 days): https://we.tl/ERCKCl4DyT (it can be viewed through any text editor)

Didn’t you say your hardware is to weak for AQ? Why?
I planed to play some test games between Zenith and AQ, but my 1080 is weaker than your 1080TI.

Nevertheless I don’t like AQ’s style I think. It played a lot of useless ataris in the two games I played against it. At least I lost my interest in AQ, but maybe I should give it a second chance.

Edit: Oh, I see … you played against RoyalLeela.

Maybe I didn't explain very well but 1080TI is not my hardware, it belongs to Roy who runs Leela on OGS.
And because I don't have GPU and AQ is known to play poorly without GPU. That's the catch.

Could I please ask someone with a decent computer what Zen/Leela/AQ other strong bots think (win % and does recent Leela also give points lead estimate?) of this position from a recent game by Iyama Yuta (you can click the diagram to download sgf). 6.5 komi, black has 1 capture and white none if it's not Chinese scoring so you need to know. Black resigned at this position so I'd like to know how bad the bots think it is for him and compare to my human positional judgement at forum/viewtopic.php?p=225078#p225078 where I said white leads by 3.5 points, but with an estimated error of 5-10 points.

Click Here To Show Diagram Code

[go]$$W Iyama Yuta (white) vs Yang Dingxin
$$ +---------------------------------------+
$$ | . . . . . . . . . . O . . . . . O . . |
$$ | . . . . . . . . . O X X X O X . X O . |
$$ | . . . O . O O X . O X O O O X X O . . |
$$ | . . . , O . X O . O O X X X . X O . . |
$$ | . . X X . X X O . O X X . X . X O . . |
$$ | . . . O X . O X X X O . . . O O X O . |
$$ | . . . . X . O O . X O . O O . . X O . |
$$ | . . . . . . . . . X . . . X X . X . . |
$$ | . . . X . X X . . . . O . . . . . . . |
$$ | . . . , . O X X X X O . O . . , . . . |
$$ | . . . . . O O O O X O X O . . X . . . |
$$ | . . . . . . . . X O O X . . . . . . . |
$$ | . . . . . . X O X . . X O . . . . . . |
$$ | . . . . . . X O . O O X X X . X . . . |
$$ | . . O . . O O . . . . O O . 1 . . . . |
$$ | . . O X . . . . . , . . . . O X O . . |
$$ | . X X . . X O . . . . . . . . . O . . |
$$ | . . . . X . X O X . . . . . . . . . . |
$$ | . . . . . X . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

Why ask Zen when you can ask FineArt
Move 124 B has 38.5% winrate 黑棋无功而返
Move 117 W has 67.6% winrate
Move 111 W has 68.3% winrate 白棋明显好下
Move 109 W has 66.1% winrate 黑棋现在紧张了
Move 99 W has 56.7% winrate
Move 91 B has 61.0% winrate 这样白棋大龙有点危险啊
Move 86 B has 57.9% winrate 黑棋霸王硬上弓 还是白棋危险
Move 80 B has 57.7% winrate 白棋远去了
Move 72 B has 57.8% winrate 黑棋露骨的杀棋
Move 50 B has 43.9% winrate 白棋实地比较多
Move 46 B has 42.7% winrate
Move 41 W has 58.4% winrate 喜欢白棋
Move 38 B has 46.2% winrate 大乱斗
Move 33 W has 57.8% winrate 白棋凶险异常啊
Move 32 B has 46.2% winrate
Move 31 W has 54.2% winrate 感觉白棋不错啊
And so on... you don't need opening sequence winrate right?

Are those FineArt win % from Fox broadcast?

Yes

Zen would play H18 and after these upper exchanges suggests E6, leaving right side as it is. This variation leads to White 57%.
I also analyzed your sequence at move 79 as well as current position.
Thanks! Game between Ke Jie and Iyama is on on Fox. FineArt says W has a lead of 54.4% after the first move.

Edit: I updated the sgf with comparison between Zen and FineArt. Sometimes they totaly agree, and sometimes they disagree a bit. Zen thinks B's last move was a huge mistake. Which may explain why B resigned?

But wait! FineArt can also show its sequence too!!!

At least it's better than relay room in Tygem which is about 30-40 moves on the game that isn't happen yet. Look like the broadcaster has too much free time or just can't wait for the actual game.

That aside, I try to dig if there's some game that FineArt comment on move 0 so we'll know how it think about komi. The 54.4% winrate for white come from black play 3-4 opening so we can't use this number