Are kiudee parameters the best?

[Ovyron]

You probably mean 20 seconds per move, right? Or your full CPU churns out 15 nps with the fast T59 net? My phone is much faster.

Yeah, it's the slowest thing I've ever seen:

CPU Leela Openblass 256x20-mod-swa-1541000.pb network:

1/02 0:02 +0.32 1.e4 e5 (3) 0
2/03 0:07 +0.29 1.e4 e5 2.Nf3 (5)
2/04 0:12 +0.30 1.e4 e5 2.Nf3 Nc6 (8)
3/05 0:16 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 (13)
3/06 0:21 +0.22 1.d4 Nf6 2.c4 e6 3.g3 Bb4+ (20)
3/06 0:26 +0.22 1.d4 Nf6 2.c4 e6 3.g3 Bb4+ 4.Qd2 (21)
4/07 0:26 +0.21 1.d4 Nf6 2.c4 e6 3.g3 Bb4+ 4.Bd2 (33) 0
4/07 0:31 +0.21 1.d4 Nf6 2.c4 e6 3.g3 Bb4+ 4.Bd2 Bxd2+ (33) 0
4/08 0:33 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Ng4 (45) 0
4/08 0:38 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.Bxc6 (52) 0
4/09 0:40 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.Bxc6 (67) 0
5/09 0:41 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nxf2 (75) 0
5/09 0:46 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nxf2 (82) 0
5/10 0:48 +0.22 1.d4 Nf6 2.c4 e6 3.g3 Bb4+ 4.Bd2 Be7 5.Bg2 d5 (103) 0
5/10 0:48 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 (108) 0
5/10 0:53 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.c3 Nf6 5.d3 (114) 0
5/11 0:56 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 (130) 0
5/11 0:56 +0.21 1.d4 Nf6 2.c4 e6 3.g3 Bb4+ 4.Bd2 Be7 5.Nf3 O-O 6.Bg2 (157) 0
5/11 1:01 +0.21 1.d4 Nf6 2.c4 e6 3.g3 Bb4+ 4.Bd2 Be7 5.Nf3 O-O 6.Bg2 (161) 0
5/11 1:01 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.Bxf7+ (167) 0
6/12 1:03 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 O-O 6.d3 (185) 0
6/12 1:08 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 O-O 6.d3 Bxf2+ (216) 0
6/13 1:11 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.d3 O-O (246) 0
6/13 1:16 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.c3 O-O (277) 0
6/14 1:19 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.d3 O-O 7.c3 (348) 0
6/14 1:24 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.d3 O-O 7.c3 (360) 0
7/14 1:24 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.d3 O-O 7.c3 (376) 0
7/15 1:26 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.c3 O-O 7.d3 (410) 0
7/15 1:31 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.c3 O-O 7.d3 (457) 0
7/16 1:34 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.d3 O-O 7.c3 a6 (494) 0
7/16 1:39 +0.29 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.a4 h6 8.Re1 (542) 0
7/17 1:41 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.a4 h6 8.Re1 Bxf2+ (631) 0
7/17 1:46 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.a4 h6 8.Re1 Bxf2+ (644) 0
8/18 1:48 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.Re1 h6 8.a4 Ba7 (743) 0
8/18 1:53 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.Re1 h6 8.a4 Ba7 (745) 0
8/19 1:55 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.Re1 h6 8.a4 Ba7 (784) 0
8/19 2:00 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.Re1 h6 8.a4 Ba7 (800) 0
8/20 2:03 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.Re1 h6 8.a4 Ba7 (822) 0
8/20 2:08 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.Re1 h6 8.a4 Ba7 (840) 0
8/21 2:10 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.Re1 h6 8.a4 Ba7 (880) 0
8/21 2:15 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (897) 0
8/21 2:20 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (912) 0
8/22 2:24 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (996) 0
8/22 2:29 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.009) 0
8/23 2:33 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.050) 0
9/23 2:35 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.138) 0
9/23 2:40 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.154) 0
9/24 2:41 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.180) 0
9/24 2:46 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.274) 0
9/25 2:49 +0.26 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.370) 0
9/25 2:54 +0.26 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.406) 0
9/26 2:57 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.434) 0
9/26 3:02 +0.26 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.577) 0
9/26 3:07 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.624) 0
9/27 3:08 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.678) 0
10/27 3:09 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.711) 0
10/27 3:14 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.778) 0
10/28 3:18 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.823) 0
10/28 3:23 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.926) 0
10/28 3:28 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.002) 0
10/29 3:29 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.056) 0
10/29 3:34 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.167) 0
10/29 3:39 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.236) 0
10/29 3:44 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.308) 0
10/29 3:49 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.497) 0
10/29 3:54 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.591) 0
10/29 3:59 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.646) 0
10/29 4:04 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.646) 0
11/29 4:05 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.776) 0
11/29 4:10 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.871) 0
11/29 4:15 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.871) 0
11/29 4:20 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 (3.019) 0
11/29 4:25 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.230) 0
11/29 4:30 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.337) 0
11/29 4:35 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.337) 0
11/29 4:40 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.337) 0
11/30 4:42 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.462) 0
11/30 4:47 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.565) 0
11/30 4:52 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.708) 0
11/31 4:54 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.766) 0
11/31 4:59 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.890) 0
11/31 5:04 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.890) 0
11/31 5:06 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.890) 0

Luckily after this initial run it reaches nodes much faster, so I average some 11k in 11 minutes.

Isn't this how it's suppose to be? Can you post Leela's output on your phone to compare? Because if it's 600 times faster perhaps I've been running it on the wrong hardware. A Quadcore, pfft!

[corres]

...
If you want to tune yourself the main one imo. To start with is temperature policy. And go from there like cput. This widens or narrows the search. I would not get caught up in any type of node counting. Why Kuidee and other settings work is because it allows Lc0 to search deeper with the same number of nodes. The trick is to do this deeper search without missing tactics. That is what you are tuning.

The main aim of kiudee settings is to tighten and to deepen the search of Leela. This obviously gives benefit to find positions with higher evaluation value what is caused by using unbalanced opening positions and/or tactical opportunity. So the effectiveness of kiudee settings depend on not only the speed of Leela but the types of the game too.

I agree, tuning the NN needs to be done to your system setup. and needs. But there are good general settings better than default. I always suggest starting with temp policy.You can tune very well with just this one setting.

If somebody needs every each Elo from Leela that one really need to tune some parameters of Leela - even tuning the kuidee also. But in practice it is enough to modify parameters of Leela from GTX (Default Leela parameters) to RTX if one uses RTX card. These parameters are: backend=cudnn-fp16 and NNCachesize=(5-10)x1000000.
It might need more modification if one uses more than one powerful RTX cards like CCC or TCEC.

[mwyoung]

...
If you want to tune yourself the main one imo. To start with is temperature policy. And go from there like cput. This widens or narrows the search. I would not get caught up in any type of node counting. Why Kuidee and other settings work is because it allows Lc0 to search deeper with the same number of nodes. The trick is to do this deeper search without missing tactics. That is what you are tuning.

The main aim of kiudee settings is to tighten and to deepen the search of Leela. This obviously gives benefit to find positions with higher evaluation value what is caused by using unbalanced opening positions and/or tactical opportunity. So the effectiveness of kiudee settings depend on not only the speed of Leela but the types of the game too.

I agree, tuning the NN needs to be done to your system setup. and needs. But there are good general settings better than default. I always suggest starting with temp policy.You can tune very well with just this one setting.

If somebody needs every each Elo from Leela that one really need to tune some parameters of Leela - even tuning the kuidee also. But in practice it is enough to modify parameters of Leela from GTX (Default Leela parameters) to RTX if one uses RTX card. These parameters are: backend=cudnn-fp16 and NNCachesize=(5-10)x1000000.
It might need more modification if one uses more than one powerful RTX cards like CCC or TCEC.

Kuidee is just a name of a tune of Lc0. And you don't need to do anything. But lets do a little thought experiment. What is Lc0 best settings if you had a infinitely fast computer system. Remember Lc0 is not a A/B search. It is a MCTS. So the best settings would be the widest possible search, with the most roll outs. Since none of us have a infinitely fast system. We must narrow the search, or we search too shallow, and miss tactics. And if we go too far and narrow too much we have the same problem. Meaning that Default would work fine, at certain time controls, and system speeds. But you leave much on the table for Lc0 in terms of Elo depending on System and Time Controls. Remember hardware speed = time.

[corres]

...
If you want to tune yourself the main one imo. To start with is temperature policy. And go from there like cput. This widens or narrows the search. I would not get caught up in any type of node counting. Why Kuidee and other settings work is because it allows Lc0 to search deeper with the same number of nodes. The trick is to do this deeper search without missing tactics. That is what you are tuning.

The main aim of kiudee settings is to tighten and to deepen the search of Leela. This obviously gives benefit to find positions with higher evaluation value what is caused by using unbalanced opening positions and/or tactical opportunity. So the effectiveness of kiudee settings depend on not only the speed of Leela but the types of the game too.

I agree, tuning the NN needs to be done to your system setup. and needs. But there are good general settings better than default. I always suggest starting with temp policy.You can tune very well with just this one setting.

If somebody needs every each Elo from Leela that one really need to tune some parameters of Leela - even tuning the kuidee also. But in practice it is enough to modify parameters of Leela from GTX (Default Leela parameters) to RTX if one uses RTX card. These parameters are: backend=cudnn-fp16 and NNCachesize=(5-10)x1000000.
It might need more modification if one uses more than one powerful RTX cards like CCC or TCEC.

Kuidee is just a name of a tune of Lc0. And you don't need to do anything. But lets do a little thought experiment. What is Lc0 best settings if you had a infinitely fast computer system. Remember Lc0 is not a A/B search. It is a MCTS. So the best settings would be the widest possible search, with the most roll outs. Since none of us have a infinitely fast system. We must narrow the search, or we search too shallow, and miss tactics. And if we go too far and narrow too much we have the same problem. Meaning that Default would work fine, at certain time controls, and system speeds. But you leave much on the table for Lc0 in terms of Elo depending on System and Time Controls. Remember hardware speed = time.

So if I run a test with default RTX settings and an another test with kiudee settings and there is no great difference between them I can use the default RTX settings without important loss.

[mwyoung]

...
If you want to tune yourself the main one imo. To start with is temperature policy. And go from there like cput. This widens or narrows the search. I would not get caught up in any type of node counting. Why Kuidee and other settings work is because it allows Lc0 to search deeper with the same number of nodes. The trick is to do this deeper search without missing tactics. That is what you are tuning.

The main aim of kiudee settings is to tighten and to deepen the search of Leela. This obviously gives benefit to find positions with higher evaluation value what is caused by using unbalanced opening positions and/or tactical opportunity. So the effectiveness of kiudee settings depend on not only the speed of Leela but the types of the game too.

I agree, tuning the NN needs to be done to your system setup. and needs. But there are good general settings better than default. I always suggest starting with temp policy.You can tune very well with just this one setting.

If somebody needs every each Elo from Leela that one really need to tune some parameters of Leela - even tuning the kuidee also. But in practice it is enough to modify parameters of Leela from GTX (Default Leela parameters) to RTX if one uses RTX card. These parameters are: backend=cudnn-fp16 and NNCachesize=(5-10)x1000000.
It might need more modification if one uses more than one powerful RTX cards like CCC or TCEC.

Kuidee is just a name of a tune of Lc0. And you don't need to do anything. But lets do a little thought experiment. What is Lc0 best settings if you had a infinitely fast computer system. Remember Lc0 is not a A/B search. It is a MCTS. So the best settings would be the widest possible search, with the most roll outs. Since none of us have a infinitely fast system. We must narrow the search, or we search too shallow, and miss tactics. And if we go too far and narrow too much we have the same problem. Meaning that Default would work fine, at certain time controls, and system speeds. But you leave much on the table for Lc0 in terms of Elo depending on System and Time Controls. Remember hardware speed = time.

So if I run a test with default RTX settings and an another test with kiudee settings and there is no great difference between them I can use the default RTX settings without important loss.

Yes! I only switched to Kiudee because it had better results a fast time controls then my own tuned settings ON MY SYSTEM. At longer time controls there was much less difference. But we have fast systems. If you are running on a slower system. Kuidee would have more of a impact at longer time controls on a slower system.

[Laskos]

You probably mean 20 seconds per move, right? Or your full CPU churns out 15 nps with the fast T59 net? My phone is much faster.

Yeah, it's the slowest thing I've ever seen:

CPU Leela Openblass 256x20-mod-swa-1541000.pb network:

1/02 0:02 +0.32 1.e4 e5 (3) 0
2/03 0:07 +0.29 1.e4 e5 2.Nf3 (5)
2/04 0:12 +0.30 1.e4 e5 2.Nf3 Nc6 (8)
3/05 0:16 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 (13)
3/06 0:21 +0.22 1.d4 Nf6 2.c4 e6 3.g3 Bb4+ (20)
3/06 0:26 +0.22 1.d4 Nf6 2.c4 e6 3.g3 Bb4+ 4.Qd2 (21)
4/07 0:26 +0.21 1.d4 Nf6 2.c4 e6 3.g3 Bb4+ 4.Bd2 (33) 0
4/07 0:31 +0.21 1.d4 Nf6 2.c4 e6 3.g3 Bb4+ 4.Bd2 Bxd2+ (33) 0
4/08 0:33 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Ng4 (45) 0
4/08 0:38 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.Bxc6 (52) 0
4/09 0:40 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.Bxc6 (67) 0
5/09 0:41 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nxf2 (75) 0
5/09 0:46 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nxf2 (82) 0
5/10 0:48 +0.22 1.d4 Nf6 2.c4 e6 3.g3 Bb4+ 4.Bd2 Be7 5.Bg2 d5 (103) 0
5/10 0:48 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 (108) 0
5/10 0:53 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.c3 Nf6 5.d3 (114) 0
5/11 0:56 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 (130) 0
5/11 0:56 +0.21 1.d4 Nf6 2.c4 e6 3.g3 Bb4+ 4.Bd2 Be7 5.Nf3 O-O 6.Bg2 (157) 0
5/11 1:01 +0.21 1.d4 Nf6 2.c4 e6 3.g3 Bb4+ 4.Bd2 Be7 5.Nf3 O-O 6.Bg2 (161) 0
5/11 1:01 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.Bxf7+ (167) 0
6/12 1:03 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 O-O 6.d3 (185) 0
6/12 1:08 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 O-O 6.d3 Bxf2+ (216) 0
6/13 1:11 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.d3 O-O (246) 0
6/13 1:16 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.c3 O-O (277) 0
6/14 1:19 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.d3 O-O 7.c3 (348) 0
6/14 1:24 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.d3 O-O 7.c3 (360) 0
7/14 1:24 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.d3 O-O 7.c3 (376) 0
7/15 1:26 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.c3 O-O 7.d3 (410) 0
7/15 1:31 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.c3 O-O 7.d3 (457) 0
7/16 1:34 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.Re1 d6 6.d3 O-O 7.c3 a6 (494) 0
7/16 1:39 +0.29 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.a4 h6 8.Re1 (542) 0
7/17 1:41 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.a4 h6 8.Re1 Bxf2+ (631) 0
7/17 1:46 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.a4 h6 8.Re1 Bxf2+ (644) 0
8/18 1:48 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.Re1 h6 8.a4 Ba7 (743) 0
8/18 1:53 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.Re1 h6 8.a4 Ba7 (745) 0
8/19 1:55 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.Re1 h6 8.a4 Ba7 (784) 0
8/19 2:00 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.Re1 h6 8.a4 Ba7 (800) 0
8/20 2:03 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.Re1 h6 8.a4 Ba7 (822) 0
8/20 2:08 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.Re1 h6 8.a4 Ba7 (840) 0
8/21 2:10 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.O-O Nf6 5.d3 d6 6.c3 a6 7.Re1 h6 8.a4 Ba7 (880) 0
8/21 2:15 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (897) 0
8/21 2:20 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (912) 0
8/22 2:24 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (996) 0
8/22 2:29 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.009) 0
8/23 2:33 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.050) 0
9/23 2:35 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.138) 0
9/23 2:40 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.154) 0
9/24 2:41 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.180) 0
9/24 2:46 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.274) 0
9/25 2:49 +0.26 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.370) 0
9/25 2:54 +0.26 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.406) 0
9/26 2:57 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.434) 0
9/26 3:02 +0.26 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.577) 0
9/26 3:07 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.624) 0
9/27 3:08 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.678) 0
10/27 3:09 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.711) 0
10/27 3:14 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.778) 0
10/28 3:18 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.823) 0
10/28 3:23 +0.27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (1.926) 0
10/28 3:28 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.002) 0
10/29 3:29 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.056) 0
10/29 3:34 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.167) 0
10/29 3:39 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.236) 0
10/29 3:44 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.308) 0
10/29 3:49 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.497) 0
10/29 3:54 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.591) 0
10/29 3:59 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.646) 0
10/29 4:04 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.646) 0
11/29 4:05 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.776) 0
11/29 4:10 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.871) 0
11/29 4:15 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (2.871) 0
11/29 4:20 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 (3.019) 0
11/29 4:25 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.230) 0
11/29 4:30 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.337) 0
11/29 4:35 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.337) 0
11/29 4:40 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.337) 0
11/30 4:42 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.462) 0
11/30 4:47 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.565) 0
11/30 4:52 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.708) 0
11/31 4:54 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.766) 0
11/31 4:59 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.890) 0
11/31 5:04 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.890) 0
11/31 5:06 +0.28 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Nxe4 5.d4 Nd6 6.Bxc6 dxc6 7.dxe5 Nf5 <snip> (3.890) 0

Luckily after this initial run it reaches nodes much faster, so I average some 11k in 11 minutes.

Isn't this how it's suppose to be? Can you post Leela's output on your phone to compare? Because if it's 600 times faster perhaps I've been running it on the wrong hardware. A Quadcore, pfft!

Hmmm.
???
Are these Nodes? Are you using BLAS backend?


My phone with T_59 net shows some 242 NPS after about 10 seconds (so almost 2500 total nodes in 10 seconds):



T40 nets are about 5 times slower, or on my phone would be about 50 NPS. What's the matter with your 10 NPS on a quad-core PC? Is it a Xeon 400 MHz of 20 years ago?

[Ovyron]

Okay, thanks, that's really weird. My output comes from a Q8200 2.33Ghz @ 4CPU, if my phone is faster it'll probably be because Leela hates my computer, but now I really need to check how she runs on my phone. A x250 speedup for running Leela on a device I already have??

[Laskos]

Okay, thanks, that's really weird. My output comes from a Q8200 2.33Ghz @ 4CPU, if my phone is faster it'll probably be because Leela hates my computer, but now I really need to check how she runs on my phone. A x250 speedup for running Leela on a device I already have??

10 NPS with 256x20b net is really weird on your CPU. I frankly have never seen such a tortoise on any of my devices. I suspect a backend or library issue.

[Ovyron]

Yeah, all nets run like this. It'd be cool if there was a problem and solving it would speed up Leela (to levels perhaps even higher than your phone!?)

Something noticeable is that it starts only using a single CPU, it takes a long while to start using the 4 threads.

Do these settings look fine? Perhaps my GUI is messing something up when installing Leela (some weirdess is going on, as RamLimitMb can only be set to 0 or negative values):

[Laskos]

Yeah, all nets run like this. It'd be cool if there was a problem and solving it would speed up Leela (to levels perhaps even higher than your phone!?)

Something noticeable is that it starts only using a single CPU, it takes a long while to start using the 4 threads.

Do these settings look fine? Perhaps my GUI is messing something up when installing Leela (some weirdess is going on, as RamLimitMb can only be set to 0 or negative values):

NNCacheSize is 200000 by default, and can be larger. Weights file is not set, it will likely take the latest saved in the same folder. Could you input UCI options and then "go" from the command prompt? Set also the desired weights file UCI option.