Chess Forum

Next example:


2r2r1k/qb3p1p/p2p1Np1/np1Pp1P1/2P5/3RP2P/5R2/4Q1K1 w - - 0 1

On 6 Cores: With 1 Networt Rf4! = immediately.

Stockfish 080124:
On 6 Cores: With 2 networks 60s needed for Rf4.

Solista wrote: ↑Tue Jan 09, 2024 12:49 pm Next example:

You could show this examples to SF team on github, maybe they could do something about it. I very doubt, but it's still worth a try.

It's very easy to change that. But I doubt the SF developers will do it. The small network brings +2 Elo in the bullet. This was important to the developers. I have now modified my engine and increased the threshold for handover to the small network in evaluate.cpp. In my new engine version (private only) both positions are now solved immediately.

Analyses of my new Leptir version:

First position:
r1n2N1k/2n2K1p/3pp3/5Pp1/b5R1/8/1PPP4/8 w - - 0 1

Analysis by Leptir 100124:

1.Ng6+ hxg6 2.Rg2 Be8+ 3.Kf8 Ra4 4.b4 Rxb4 5.c4 Rxc4 6.d4 Rxd4 7.Rh2+ Rh4 8.Rxh4+ gxh4 9.fxg6 Bf7 10.g7+ Kh7 11.g8Q+ Bxg8
Depth: 22/23 00:00:03 9873kN, tb=15846
The position is equal.

Second position:
2r2r1k/qb3p1p/p2p1Np1/np1Pp1P1/2P5/3RP2P/5R2/4Q1K1 w - - 0 1

Analysis by Leptir 100124:

1.Rf4 h5 2.Qh4 Rxc4 3.Rxc4 Rc8 4.Rcc3 Rc4 5.Rxc4 Nxc4 6.Kh2 Bc8 7.e4 b4 8.Rf3 Nd2 9.Rd3
Depth: 20/36 00:00:01 12707kN, tb=4
The position is equal.

For me, analysis mode is more important than +2 bullet Elo. That's why I modified my engine. Once I have tested everything, I will release the new engine version.

BTW:
Of course I tested the new SF 080124 with two networks with my ENET-2023.

On 6 Cores with 60s - --> Solved: 95 positions of 110 (Leptir 090124 107 of 110).

For the first position I posted here it looks like this on 6 cores:
2. ENET-2023 002 (ERET 002) - Zugzwang, Ng6+, Akerblom 1961 Solved in 24.08s/32

Pos 2:
25. ENET-2023 025 (CCC) - Ausgleich, Rf4 Solved in 60.23s/31

SF-PB-080124_x64 "Update default main net to nn-baff1edbea57.nnue"
Link replaced with working one.
https://pixeldrain.com/u/McuFsJEX

Skynet wrote: ↑Thu Jan 11, 2024 1:33 am SF-PB-080124_x64 "Update default main net to nn-baff1edbea57.nnue"
Link replaced with working one.
https://pixeldrain.com/u/McuFsJEX

Thank you!

Solista wrote: ↑Wed Jan 10, 2024 10:15 am It's very easy to change that. But I doubt the SF developers will do it. The small network brings +2 Elo in the bullet. This was important to the developers. I have now modified my engine and increased the threshold for handover to the small network in evaluate.cpp. In my new engine version (private only) both positions are now solved immediately.

Analyses of my new Leptir version:

First position:
r1n2N1k/2n2K1p/3pp3/5Pp1/b5R1/8/1PPP4/8 w - - 0 1

Analysis by Leptir 100124:

1.Ng6+ hxg6 2.Rg2 Be8+ 3.Kf8 Ra4 4.b4 Rxb4 5.c4 Rxc4 6.d4 Rxd4 7.Rh2+ Rh4 8.Rxh4+ gxh4 9.fxg6 Bf7 10.g7+ Kh7 11.g8Q+ Bxg8
Depth: 22/23 00:00:03 9873kN, tb=15846
The position is equal.

Second position:
2r2r1k/qb3p1p/p2p1Np1/np1Pp1P1/2P5/3RP2P/5R2/4Q1K1 w - - 0 1

Analysis by Leptir 100124:

1.Rf4 h5 2.Qh4 Rxc4 3.Rxc4 Rc8 4.Rcc3 Rc4 5.Rxc4 Nxc4 6.Kh2 Bc8 7.e4 b4 8.Rf3 Nd2 9.Rd3
Depth: 20/36 00:00:01 12707kN, tb=4
The position is equal.

For me, analysis mode is more important than +2 bullet Elo. That's why I modified my engine. Once I have tested everything, I will release the new engine version.

BTW:
Of course I tested the new SF 080124 with two networks with my ENET-2023.

On 6 Cores with 60s - --> Solved: 95 positions of 110 (Leptir 090124 107 of 110).

For the first position I posted here it looks like this on 6 cores:
2. ENET-2023 002 (ERET 002) - Zugzwang, Ng6+, Akerblom 1961 Solved in 24.08s/32

Pos 2:
25. ENET-2023 025 (CCC) - Ausgleich, Rf4 Solved in 60.23s/31

ShashChess-34.5 solves instantly!
But Live Book=true.

I saw that no new Stockfish-Polyglot has been published here for a long time. Yesterday there was a new Stockfish update: https://www.abrok.eu/stockfish/

Author: Robert Nurnberg @ elitebook
Date: Thu Mar 7 21:10:33 2024 +0100
Timestamp: 1709842233

Now with new network nn-1ceb1ade0001.nnue

I have implemented the polyglot book and as feature Slow Mover.

Download SF PB 070324 (avx2, avx512, bmi2, sse41, source code):
https://pixeldrain.com/u/dJonJD2i

This engine will never become my analysis engine. In the following position, the Stockfish needs 48s on 8 cores to find Qxf6+. This is the worst value of all Stockfishs in a year. Sad. Previous version only takes 2s.

Qxf6+!
1B1r4/rp2npkp/2b1pbp1/1qp5/nPN1R3/1P1P1QP1/2P2PBP/5R1K w - - 0 1

Analysis by Stockfish 070324-avx2:

1.Qxf6+ Kxf6 2.Be5+ Kg5 3.Bg7 Bxe4 4.f4+ Kh5 5.Bxe4 g5 6.Ne5 Qc6 7.Bf6 h6 8.Bxe7 Qd5 9.Bxd8 Qxe5 10.fxe5 Nc3 11.bxc5 Nxe4 12.dxe4 Ra2 13.c6 bxc6 14.Rf2 g4 15.Kg2 Kg6 16.Bb6 h5 17.h3 Ra1 18.hxg4 hxg4 19.Rf6+ Kg7 20.Rf1 Ra2 21.Rf2 Kg6 22.b4 Ra1 23.Re2 c5 24.b5 Ra3
Depth: 35/63 00:00:48 505MN, tb=12573

Engine is a total DISASTER!

It's unbelievable what Stockfish 070324 analyzes here:

Bxh7+!
1r3rk1/3bbppp/1qn2P2/p2pP1P1/3P4/2PB1N2/6K1/qNBQ1R2 w - - 0 1

Analysis by Stockfish 070324-avx2:

1.Qe1 Qaxb1 2.Bxb1 Qxb1 3.Nh4 Rfe8 4.Kh2 Qe4 5.Ng2 Bf8 6.Qxe4 dxe4 7.Ne3 Nxe5 8.dxe5 Rxe5 9.Nc4 Rc5 10.Bf4 Rb7 11.Nd2 Bc6 12.c4 g6 13.Kg3 a4 14.Ra1 e3 15.Bxe3 Bd6+ 16.Kf2 Rf5+ 17.Ke2
Depth: 33/48 00:01:37 1760MN, tb=18014
Black is clearly winning.
-+ (-5.67)

1. Bxh7 wins! Even after 3 minutes Stockfish doesn't see this. Even after 1. Bxh7+ Kxh7, Stockfish 070324 sees Black as having an advantage.

1r3r2/3bbppk/1qn2P2/p2pP1P1/3P4/2P2N2/6K1/qNBQ1R2 w - - 0 1

Analysis by Stockfish 070324-avx2:

1.Rh1+ Kg8 2.Qe1 Qbb2+ 3.Bxb2 Rxb2+ 4.Nfd2 Qxb1 5.Qxb1 Rxb1 6.Rxb1 Rb8 7.Rxb8+ Nxb8 8.fxe7 Nc6 9.e8Q+ Bxe8 10.c4 Ne7 11.cxd5 Nxd5 12.g6 a4 13.Nc4 fxg6 14.Kg3 Bb5 15.Na3 Bd3 16.Kh4 Kf7 17.Kg5 Ke7 18.Kh4 Nc7 19.Kg5 Ke6 20.Kf4 Kd5 21.Ke3 Bf5 22.Kf4 Be6 23.Ke3 g5 24.Nc2 Kc4
-+ (-5.36) Tiefe: 30/68 00:00:30 513MN, tb=472433

Crazy!

Analysis by SF-Cor 1Mar24-3072-avx2:

1.Rh1+ Kg8 2.Rh8+ Kxh8 3.Qh1+ Kg8 4.g6 Qbb2+ 5.Bxb2 Qxb2+ 6.Kg3 Qf2+ 7.Kxf2 fxg6 8.Ng5 Rxf6+ 9.exf6 Bxf6 10.Qxd5+ Kh8 11.Nf7+ Kh7 12.Qxd7 Nd8 13.Nxd8 Rxd8 14.Qh3+ Kg8 15.Nd2 a4 16.Ne4 Rf8 17.Qe6+ Kh7 18.Ke2 a3 19.Qa6 Be7 20.Qa7 Re8 21.Kd3 Kh8 22.Qd7 Ra8 23.Qb7 Re8 24.Qc6 Rb8 25.Qc7 Ra8 26.Qh2+ Kg8 27.Qa2+ Kf8
+- (4.58) Tiefe: 31/65 00:00:07 92844kN, tb=42708

SF-Cor 1Mar24-3072 say White is winning. Stockfish 070324 is a Disaster!

SF-PB-16.1_x64
Special thanks to Marco, sarona (Ron) and locutus2 (Stefan Geschwentner) for the help with code!!
https://pixeldrain.com/u/L53cvpYR