How to run KataGo in Lizzie v0.7.2?

[goame]

I have 91% GPU load running Leela Zero and
I have 86% GPU load running KataGo.

[lightvector]

I checked it with a deeper position and without symmetries.
Leela Zero is 2x faster.
That means to improve the speed to Leela Zeros level will give all KataGo nets an improvement of 50 Elo.
Even more are possible, because I don't think that Leela Zeros speed is perfect.

I think the 8x speed gain at the beginning is also very important, because the depth is much higher = better moves. Even if the opponent can find this moves too, he will have later a lower depth and the chance that he overlooks the best moves is much higher.
This 8x speed improvement gains the main benefit not in the beginning but some moves later.

Thanks, yes, 2x sounds more reasonable, about in line with how much one might imagine optimizing the GPU implementation in the future. I think you're mostly mistaken about how much the 8x matters. As I mentioned before, it should be equivalent to maybe being able to save 1% to 2% on one's time control due to tree reuse, or equivalently, 1% to 2% speedup. Nice, but not a big deal right now, particularly if it causes users to be confused about the performance of the bot sometimes.

[Bill Spight]

No, I mean the first 3 moves. It's 8x on the first move, and then after a corner is occupied, depending on 4-4 or 3-4 it will be 2x or immediately 1x. It may stay 2x if the opponent takes the diagonally opposing 4-4, and also if the game develops to all 4-4 points. Thereafter, probably symmetry is broken for that game forever.

So about for the first three moves that, say, Black, will make in a game, you can have a speedup, and for two of them, it's at best 2x (only for 4-4 points), not 8x.

Symmetry breaking is one reason behind my question about focused (or restricted) search. E. g., at move one, restrict search to the right-top-right section, Aside from possible speedup, doing so during training makes an asymmetric neural network expected. (To train the bot to make move one in that section it is probably best to allow search anywhere, but to penalize choices outside it.) Whether that matters is another question. I didn't think that 2x symmetry at move two or later would make much of a difference.