The Great Stockfish NPS Debate

[dkappe]

With the Stockfish nets getting ever bigger, a small but vocal minority has voiced concern that a drop in NPS could somehow harm its fitness as a chess engine. Maybe it is vulnerable to being “outsearched” by a strong, nimble competitor.

I decided to run a test with SF10 against its successors to see if the slower ones do suffer from some sort of horizon effect and fall off a cliff. The conclusion? Nope. The bigger the net, the more resistant the SF version seems to be to being outsearched.



The full article can be found here. https://github.com/dkappe/leela-chess-w ... NPS-Debate

[Sopel]

Another nice side-effect is that the TT becomes more efficient and requires less TB hits.

[yurikvelo]

Offtopic: I run SF-latest (abrok) in self-play from startpos (no book), fixed time per move (about 6 billiion nodes per move).
And it played 1.c4 e5 2. g3 Nf6

For many years, SF preferred e4 or d4.
It is first time I see c4

This is horizon effect, or new nets really prefer c4 + g3?
In chessdb.cn rating c4+g5 and especially 2. .. Nf5 are rated lower than other moves

[Stephen Ham]

Hi yurikvelo,

I wonder about your result, since you're surely using a multi-core PC. I assume that non-determinism still reigns, even with NNUE engines. Right? So, if you ran a second test under the same conditions, you'd probably get a different result. After all, equal positions are equal. So to a chess engine, one equal position is as good as another.

All the best,
-Steve-

[jhellis3]

Yep, due to exponential growth of game tree and TTD. More accurate eval is very, very valuable.

EDIT: That isn't to say depth isn't valuable... it is, but progressively harder to obtain.

[yurikvelo]

I wonder about your result, since you're surely using a multi-core PC

[pgn][Date "2021.08.26"]
[Result "1/2-1/2"]
[ECO "A20"]
[Opening "English"]
[Time "22:14:14"]
[Variation "KID, 2.g3 Nf6 3.Bg2 Bc5"]
[TimeControl "0+1800"]
[PlyCount "72"]

1. c4 {(1.c4 e6 2.Nc3 d5 3.d4 Nf6 4.cxd5 exd5 5.Bg5 Be7 6.e3 h6 7.Bh4 c6
8.Qc2 0-0 9.Bd3 Re8 10.Nge2 a5 11.0-0 Nbd7 12.a3 b5 13.Nb1 Bb7 14.Nd2 Nb6
15.Rfe1 a4 16.Nf3 Ne4 17.Bxe7 Qxe7 18.Ng3 Rac8 19.Nd2 c5 20.Bxb5 cxd4
21.Qd1 dxe3 22.Ndxe4 dxe4) +0.31/43 1800} e5 {(1. ... e5 2.g3) -0.21/45
1799} 2. g3 {(2.g3 Nf6 3.Bg2 Bc5 4.Nc3 c6 5.e3 Bb6 6.d4 exd4 7.exd4 0-0
8.Nge2 d5 9.cxd5 Nxd5 10.Nxd5 cxd5 11.0-0 Bg4 12.Qd3 Nc6 13.Be3 Bxe2
14.Qxe2 Nxd4 15.Bxd4 Bxd4 16.Rad1 Qb6 17.Rd3 Qxb2 18.Qxb2 Bxb2 19.Rb1 Bf6
20.Rxb7 Rfb8 21.Rdb3 Rd8 22.h4 g6 23.h5 Rd6 24.hxg6 hxg6 25.Rd3 Re8 26.Rxa7
Re1+ 27.Bf1 Bd8 28.Rf3 Rf6 29.Rxf6 Bxf6 30.Kg2 Rd1 31.a4) +0.16/48 1799}
Nf6 {(2. ... Nf6 3.Bg2) -0.33/46 1799} 3. Bg2 {(3.Bg2 Bc5) +0.11/50 1799}
Bc5 {(3. ... Bc5 4.d3) -0.19/46 1799} 4. d3 {(4.d3 Nc6) +0.09/47 1799} O-O
{(4. ... 0-0) -0.10/45 1799} 5. Nc3 {(5.Nc3 Nc6 6.h3 Re8 7.Bg5 Bb4 8.Rc1 h6
9.Bxf6 Qxf6 10.e3 Nd4 11.Nge2 Nxe2 12.Qxe2 d6 13.0-0 c6 14.Rfd1 Qe7 15.a3
Bxc3 16.Rxc3 a5 17.d4 a4 18.dxe5 dxe5 19.Rcd3 Qc5 20.Rd6 Be6 21.Bf1 Rab8
22.Qc2 b5 23.R6d2 bxc4 24.Qxa4 c3 25.Rc1 cxb2 26.Rxc5 b1Q) +0.20/49 1800}
Nc6 {(5. ... Nc6) -0.09/45 1799} 6. a3 {(6.a3 a5 7.e3 d6 8.Nge2 Ne7 9.0-0
Ba7 10.e4 c6 11.h3 Be6 12.b3 Ne8 13.d4 f5 14.Be3 fxe4 15.Nxe4 Nf5 16.Bg5
Qc7 17.d5 Bf7 18.g4 Nd4 19.Rc1 Bg6 20.N2c3 Nf6 21.Nxf6+ gxf6 22.Bh6 f5
23.Kh2 Kh8 24.f4 Qe7 25.fxe5 dxe5 26.Bxf8 Rxf8 27.gxf5 Nxf5 28.d6 Qg7
29.Ne4 Be3 30.Rc3 Bf4+ 31.Kh1 Rd8 32.Rd3) +0.12/47 1799} d6 {(6. ... d6
7.e3 a5 8.Nge2 Ba7 9.0-0 Ne7 10.d4 Re8 11.h3 Ng6 12.d5 Bd7 13.b4 e4 14.Nf4
Ne5 15.Nxe4 Nxe4 16.Bxe4 g5 17.Nd3 Nxc4 18.Bg2 axb4 19.Qb3 b5 20.Nxb4 Qf6
21.Ra2 Bc5 22.Nc6 Qh6 23.a4 Na5 24.Nxa5 Rxa5 25.axb5 Rxa2 26.Qxa2 Bxh3
27.Qa5 Bxg2 28.Kxg2 Qg6 29.Qxc7 Qe4+ 30.Kg1 Qxd5 31.Qc6 Qxc6 32.bxc6)
-0.06/45 1800} 7. e3 {(7.e3 a5 8.Nge2 Ne7 9.0-0 Ba7 10.e4 c6 11.h3 Be6
12.b3 Ne8 13.d4 f5 14.Be3 fxe4 15.Nxe4 Nf5 16.Bg5 Qc7 17.d5 Bf7 18.N4c3 h6
19.g4 hxg5 20.gxf5 Nf6 21.Ne4 Nxe4 22.Bxe4 cxd5 23.cxd5 Qd8 24.Ng3 Bd4
25.Rb1 Be8 26.Qg4 Bd7 27.Rbc1 Bb2 28.Rc2 Bxa3 29.Kh1 Be8 30.Rd1 a4 31.bxa4
Bxa4) +0.15/47 1799} a5 {(7. ... a5) -0.16/48 1800} 8. Nge2 {(8.Nge2 Ba7
9.0-0 Ne7 10.d4 Re8 11.h3 Ng6 12.Kh2 e4 13.Nf4 c6 14.d5 Bf5 15.Nxg6 hxg6
16.b3 cxd5 17.Nxd5 a4 18.bxa4 Nxd5 19.Qxd5 Re5 20.Qd2 Rc5 21.Bb2 Rxc4
22.Rac1 Rxa4 23.Rfd1 Bb8 24.Qc3 Qf8 25.Rd4 Rxd4 26.Qxd4 Ba7 27.Qd5 Bc5
28.Bxe4 Be6 29.Qg5 Bxa3 30.Rc2 Bxb2 31.Rxb2 d5 32.Bxd5 Bxd5 33.Qxd5 Rd8
34.Qg2 Rd1 35.Rxb7) +0.20/47 1799} Ne7 {(8. ... Ne7 9.0-0 Ba7 10.h3 Be6
11.e4 c6 12.b3 Ne8 13.d4 f5 14.Be3 fxe4 15.Nxe4 Nf5 16.Bg5 Nf6 17.d5 cxd5
18.cxd5 Bd7 19.g4 Nd4 20.N2g3 Qb6 21.Bxf6 gxf6 22.b4 Kh8 23.bxa5 Qxa5
24.Rb1 Bb5 25.Re1 Qd8 26.Nf5 Bd7 27.Nfxd6 f5 28.Nxf5 Nxf5 29.gxf5 Bxf5
30.Rxb7 Qh4 31.Kh2 Bc5 32.Rb3 Bxe4 33.Rxe4 Qxf2 34.Rg3 Rxa3 35.Rxa3 Bxa3)
-0.15/46 1799} 9. d4 {(9.d4) +0.19/45 1800} Ba7 {(9. ... Ba7) -0.06/46
1800} 10. h3 {(10.h3 c6) +0.03/45 1799} Be6 {(10. ... Be6 11.b3 c6 12.0-0
Nf5 13.d5 cxd5 14.Nxd5 Nxd5 15.Bxd5 Ne7 16.Bxe6 fxe6 17.Nc3 Nf5 18.Kh2 Qb6
19.Nb5 Qc6 20.Qe2 Bb6 21.Bb2 e4 22.b4 d5 23.bxa5 dxc4 24.a4 Rxa5 25.Rfc1
Rc8 26.Kg1 Raa8 27.Rc2 h6 28.h4 Kh7 29.Rcc1 Kg8 30.Rc2) 0.00/48 1799} 11.
b3 {(11.b3 c6 12.0-0 Nf5 13.Qd3 exd4 14.exd4 h6 15.d5 cxd5 16.Nxd5 Nxd5
17.Bxd5 Bxd5 18.Qxd5 Re8 19.Qxf5 Rxe2 20.Qf3 Rc2 21.Qxb7 a4 22.Bf4 Qc8
23.Qxc8+ Rxc8 24.bxa4 R8xc4 25.Bxd6 Rxa4 26.Bb4 Bc5 27.Bxc5 Rxc5 28.Rfe1
Kh7 29.Re3 Rca5 30.Kf1 Ra7 31.Ke2 R4a6 32.Kd1 h5 33.Rf3 g5 34.Kd2 Kg7
35.Ra2 Kh7 36.Re3 Kg6 37.Ke2 Ra5 38.Kf1 Ra4 39.Rb3 f6 40.Rf3 Kg7 41.Ra1
Kg6) +0.10/56 1799} c6 {(11. ... c6 12.0-0 Nf5 13.Qd3 exd4 14.exd4 h6 15.d5
cxd5 16.Nxd5 Nxd5 17.Bxd5 Bxd5 18.Qxd5 Re8 19.Qxf5 Rxe2 20.Qf3 Rc2 21.Qxb7
a4 22.Bf4 Qc8 23.Qxc8+ Rxc8 24.bxa4 R8xc4 25.a5 Ra4 26.Bxd6 Bc5 27.Bxc5
Rxc5 28.Rfe1 Kh7 29.a6 Rca5 30.Reb1 Rxa6 31.Rb3 Kg6 32.Rf3 f6 33.Kf1 R4a5
34.Ke2 Rh5 35.h4 Rha5 36.Rc3 Ra4 37.Rd3 Ra7 38.Ke1 Kh5 39.Kf1 Kg6 40.Rc3 h5
41.Rf3 R4a5 42.Rb3 R5a6 43.Kg2 Ra5 44.a4 Rxa4) -0.02/57 1799} 12. O-O
{(12.0-0 Nf5 13.Qd3 exd4 14.exd4 h6 15.d5 cxd5 16.Nxd5 Nxd5 17.Bxd5 Bxd5
18.Qxd5 Re8 19.Qxf5 Rxe2 20.Qf3 Rc2 21.Qxb7 a4 22.Bf4 Qc8 23.Qxc8+ Rxc8
24.bxa4 R8xc4 25.a5 Ra4 26.Bxd6 Bc5 27.Bxc5 Rxc5 28.Rfe1 Kh7 29.a6 Rca5
30.Reb1 Rxa6 31.Rb3 Kg6 32.Rf3 h5 33.Ra2 Ra7 34.Kg2 f6 35.Rc3 Kf7 36.Rb3
Kg6 37.Rf3 R7a5 38.Kf1 Ra6 39.Ke2 h4 40.g4 Kh6 41.Kd2 Kg5 42.Kc2 Rf4
43.Rxf4 Kxf4 44.a4) +0.02/56 1799} Nf5 {(12. ... Nf5 13.Qd3 h6 14.d5 cxd5
15.Nxd5 Nxd5 16.Bxd5 Bxd5 17.Qxd5 Ne7 18.Qb5 Qc8 19.Kg2 h5 20.Rd1 Bc5
21.Bd2 Qf5 22.b4 Qe4+ 23.Kg1 axb4 24.axb4 Rxa1 25.Rxa1 Bd4 26.Re1 Qd3
27.Nxd4 exd4 28.exd4 Nf5 29.Be3 Qc3 30.Rc1 Qd3 31.Re1) 0.00/48 1799} 13.
Qd3 {(13.Qd3 h6 14.d5 cxd5 15.Nxd5 Nxd5 16.Bxd5 Bxd5 17.Qxd5 Ne7 18.Qb5 Qc8
19.Kh2 Qf5 20.Ra2 Bc5 21.Nc3 Qd3 22.Ne2) 0.00/48 1800} h6 {(13. ... h6
14.Ra2 exd4 15.exd4 d5 16.cxd5 Nxd5 17.Nxd5 cxd5 18.a4 Re8 19.Ba3 Rc8
20.Rd2 Qd7 21.Nf4 Nxd4 22.Rfd1 Qc7 23.Kh2 Bf5 24.Nxd5 Bxd3 25.Nxc7 Rxc7
26.Rxd3 Rc2 27.R1d2 Re2 28.Bc1 Rxf2 29.Rxc2 Rxc2 30.Be3 Nf3+ 31.Kh1 Bxe3
32.Bxf3 Rc1+ 33.Kg2 Rc2+) 0.00/48 1799} 14. Kh2 {(14.Kh2 exd4 15.exd4 d5
16.cxd5 Nxd5 17.Nxd5 cxd5 18.Bf4 Rc8 19.Rac1 Qd7 20.Rfe1 Bb6 21.Rxc8 Rxc8
22.Be5 f6 23.Bf4 g5 24.Bd2 Kg7 25.h4 Re8 26.h5 Qf7 27.Bh3 Qxh5 28.g4 Qh4
29.Ng3 Nd6 30.Nh5+ Kf8 31.Qg6 Qxf2+ 32.Bg2 Qh4+ 33.Bh3) 0.00/50 1800} exd4
{(14. ... exd4 15.exd4 d5 16.cxd5 Nxd5 17.Nxd5 cxd5 18.Bf4 Qd7 19.Rfe1 Rac8
20.h4 Rc6 21.Rac1 Re8 22.Be5 Ne7 23.Qd2 Ng6 24.Nf4 Nxf4 25.Bxf4 Rec8
26.Rxc6 Qxc6 27.Qxa5 Qc3 28.Qxc3 Rxc3 29.Bh3 Bxd4 30.Bxe6 fxe6 31.Rxe6 Bxf2
32.Rd6 Rxb3 33.Rxd5 Rxa3 34.Rd7 Rb3 35.Be5 g5 36.hxg5 hxg5 37.Kh3 Kf8
38.Kg4 Rb4+ 39.Kxg5 Rb5 40.Kf6 Rxe5) 0.00/57 1799} 15. exd4 {(15.exd4 d5
16.cxd5 Nxd5 17.Nxd5 cxd5 18.Bf4 Qd7 19.Rfe1 Rac8 20.h4 Rc6 21.Rac1 Re8
22.Be5 Ne7 23.Qd2 Ng6 24.Nf4 Nxf4 25.Bxf4 Rec8 26.Rxc6 Qxc6 27.Qxa5 Qc3
28.Qxc3 Rxc3 29.Bh3 Bxd4 30.Bxe6 fxe6 31.Rxe6 Bxf2 32.Rd6 Rxb3 33.Rxd5 Rxa3
34.Rd7 Rb3 35.Be5 Re3 36.Bxg7 Bxg3+ 37.Kg2 Bf4 38.Rxb7 Rg3+ 39.Kf2 Rxg7)
0.00/60 1799} d5 {(15. ... d5 16.cxd5 Nxd5 17.Nxd5 cxd5 18.Bf4 Qd7 19.Rac1
Rac8 20.Rfe1 Bb6 21.h4 Rxc1 22.Rxc1 Re8 23.Re1 Ne7 24.Nc3 Rc8 25.h5 Nf5
26.Nb5 Ne7) 0.00/61 1800} 16. cxd5 {(16.cxd5 Nxd5 17.Nxd5 cxd5 18.Bf4 Qd7
19.Rfe1 Rac8 20.Rac1 Bb6 21.h4 Rxc1 22.Rxc1 Re8 23.Re1 Ne7 24.Nc3 Rc8 25.h5
Nc6 26.Be3 Rd8 27.Rc1 Ba7 28.Bf3 Bf5 29.Qd2 Qe6 30.Kg1 Qe7 31.Nb5 Bb6
32.Rd1 Qf6 33.Bf4 Qe7 34.Re1 Qd7 35.Kh2 Be4 36.Bd1 Bf5) 0.00/61 1799} Nxd5
{(16. ... Nxd5 17.Nxd5 cxd5 18.Bf4 Qd7 19.Rfe1 Rac8 20.Rac1 Bb6 21.h4 Rxc1
22.Rxc1 Re8 23.Re1 Ne7 24.Nc3 Rc8 25.h5 Nc6 26.Be3 Rd8 27.Rc1 Ba7 28.Qd2
Qe7 29.Nb5 Bb6 30.Bf3 Bf5 31.Kg2 Re8 32.Kg1 Rd8 33.Rd1 Qf6 34.Bf4 Qe7
35.Re1 Qd7 36.Kh2 Be6 37.Bd1 Bf5 38.Qe3 Be6) 0.00/63 1799} 17. Nxd5
{(17.Nxd5 cxd5 18.Bf4 Qd7 19.Rfe1 Rac8 20.Rac1 Bb6 21.h4 Rxc1 22.Rxc1 Re8
23.h5 Ne7 24.Nc3 Rc8 25.Re1 Nc6 26.Be3 Rd8 27.Rc1 Ba7 28.Bf3 Bf5 29.Qd2 Be6
30.Re1 Qe7 31.Nb5 Bb8 32.Kg1 Bd6 33.Nxd6 Qxd6 34.b4 axb4 35.axb4 Qxb4
36.Qxb4 Nxb4 37.Rb1 Nc2 38.Rxb7 Ra8 39.Rb5 Ra1+ 40.Kh2 Ra2 41.Kg1) 0.00/62
1799} cxd5 {(17. ... cxd5 18.Bf4 Rc8 19.Rac1 Rc6 20.Be5 Ne7 21.b4 Rxc1
22.Rxc1 axb4 23.axb4 Nc6 24.Bf4 Qb6 25.Be3 Qxb4 26.Rb1 Qe7 27.Qb5 Rb8
28.Bxd5 Bxd5 29.Qxd5 Qf6 30.Qe4 Rd8 31.Rxb7 Bxd4 32.Rd7 Rxd7 33.Qe8+ Kh7
34.Qxd7 Bxe3 35.Qd3+ Qg6 36.Qxe3 Qf5 37.Nd4 Nxd4 38.Qxd4 Qf3 39.Kg1 g6
40.Qd2 Kg7 41.Qb2+ Kg8 42.h4 g5 43.Qd4 Kh7 44.hxg5 hxg5 45.Qe3 Qxe3
46.fxe3) 0.00/59 1800} 18. a4 {(18.a4 Rc8 19.Ba3 Re8 20.Rac1 Qd7 21.Rfe1
Bb8 22.Nc3 Red8 23.Kg1 Bd6 24.Bxd6 Qxd6 25.Nxd5 Rxc1 26.Rxc1 Nxg3 27.Qxg3
Bxd5 28.Qxd6 Rxd6 29.Bxd5 Rxd5 30.Rc7 Rxd4 31.Rxb7 g5 32.Rb5 Rd3 33.Rxa5
Rxb3 34.h4 gxh4 35.Rh5 Kg7 36.Rxh4 Ra3 37.Rb4 h5 38.Kg2 Kg6 39.f3 Ra2+
40.Kg3 f6 41.Rb5 Rxa4 42.Rd5 h4+ 43.Kh3 f5 44.f4 Rxf4) 0.00/54 1800} Rc8
{(18. ... Rc8 19.Ba3 Re8 20.Rac1 Qd7 21.Rfe1 Bb8 22.Nc3 Red8 23.Kg1 Bd6
24.Bxd6 Qxd6 25.Nxd5 Rxc1 26.Rxc1 Nxg3 27.Qxg3 Bxd5 28.Qxd6 Rxd6 29.Bxd5
Rxd5 30.Rc7 Rxd4 31.Rxb7 g5 32.Rb5 Rd3 33.Rxa5 Rxb3 34.h4 gxh4 35.Rh5 Kg7
36.Rxh4 Ra3 37.Rb4 h5 38.Kg2 Kg6 39.f3 Ra2+ 40.Kg3 f6 41.Rb5 Rxa4 42.Rd5
h4+ 43.Kh3 f5 44.f4 Rxf4) 0.00/55 1800} 19. Ba3 {(19.Ba3 Re8 20.Rac1 Qd7
21.Kg1 Rc6 22.Qd2 Bb6 23.h4 Rec8 24.Rxc6 bxc6 25.Bc5 Bxc5 26.dxc5 d4 27.Qd3
Qe7 28.Rc1 Rb8 29.Qd2 Rxb3 30.Nxd4 Nxd4 31.Qxd4 Rb4 32.Qd2 Qd7 33.Qxd7 Bxd7
34.Rd1 Be6 35.Bxc6 Rc4 36.Rd8+ Kh7 37.Bd7 Rc1+ 38.Kh2 Bb3 39.Rb8 Bd5 40.g4
Rxc5 41.Rb5 Rxb5 42.axb5 a4 43.Bc6 Bxc6 44.bxc6 a3 45.c7 a2 46.h5 a1Q
47.c8Q Qe1 48.Kg3 Qe5+ 49.Kf3 Qd5+ 50.Ke3 Qe5+ 51.Kf3) 0.00/54 1799} Re8
{(19. ... Re8 20.Rac1 Qd7 21.Kg1 Rc6 22.Qd2 Bb6 23.h4 Rec8 24.Rxc6 bxc6
25.Bc5 Qa7 26.Qc3 Qb7 27.Nf4 Rb8 28.Rb1 Bxc5 29.dxc5 Qb4 30.Qxb4 Rxb4
31.Nd3 Rb8 32.b4 Nd4 33.Rb2 g5 34.bxa5 Ra8 35.Rb4 Ne2+ 36.Kh2 Rxa5 37.Bh3
gxh4 38.Rxh4 Nc3 39.Bxe6 fxe6 40.Rxh6 Nxa4 41.Rxe6 Nxc5 42.Nxc5 Rxc5 43.f4
Kf7 44.f5 d4 45.Rd6 Rxf5 46.Rxd4) 0.00/60 1799} 20. Rac1 {(20.Rac1 Qd7
21.Kg1 Rc6 22.Qd2 Bb6 23.h4 Rec8 24.Rxc6 bxc6 25.Bc5 Qa7 26.Qc3 Qb7 27.Nf4
Rb8 28.Rb1 Bxc5 29.dxc5 Qb4 30.Qxb4 Rxb4 31.Nd3 Rb8 32.b4 Nd4 33.Rb2 g5
34.bxa5 Ra8 35.Rb4 Ne2+ 36.Kh2 Rxa5 37.Bh3 gxh4 38.Rxh4 Nc3 39.Bxe6 fxe6
40.Rxh6 Nxa4 41.Rxe6 Nxc5 42.Nxc5 Rxc5 43.f4 Kf7 44.f5 d4 45.Rd6 Rxf5
46.Rxd4) 0.00/60 1557} Qd7 21. Rxc8 Rxc8 22. Rc1 Rxc1 23. Bxc1 Bb6 24. Nf4
g5 25. Nxe6 fxe6 26. Bb2 Qf7 27. Kg1 Qf6 28. Qb5 Bxd4 29. Bxd4 Nxd4 30.
Qxa5 Ne2+ 31. Kh2 Qxf2 32. Qd8+ Kg7 33. Qe7+ Kh8 34. Qd8+ Kg7 35. Qe7+ Kh8
36. Qd8+ Kg7 {3-fold repetition} 1/2-1/2
[/pgn]

[yurikvelo]

did multiPV=4 @ startpos up to D=53 (275 BN), 16 Gb hash, Syzygy-6 (got 1650 hits at d53)

[Cornfed]

Offtopic: I run SF-latest (abrok) in self-play from startpos (no book), fixed time per move (about 6 billiion nodes per move).
And it played 1.c4 e5 2. g3 Nf6

For many years, SF preferred e4 or d4.
It is first time I see c4

This is horizon effect, or new nets really prefer c4 + g3?

Total wildshot guess - but could it be that demonstrably weaker positions can be gained with the former, but the structures obtained from the latter are less prone to creating imbalances from the Black side....thus more steady and White keeps a tiny edge longer?

I say that still half asleep...but knowing for 20 yrs I maintained a rating north of 2000 USCF with little preparation yet regular play and adopting that approach with White.

[syzygy]

Offtopic: I run SF-latest (abrok) in self-play from startpos (no book), fixed time per move (about 6 billiion nodes per move).
And it played 1.c4 e5 2. g3 Nf6

For many years, SF preferred e4 or d4.
It is first time I see c4

This is horizon effect, or new nets really prefer c4 + g3?

A horizon effect would mean that the search manages to push a tactical trick behind the search horizon, which seems rather unlikely here.

[Stephen Ham]

Hi yurikvelo,

Wonderful graph! Thank you!

My interpretation of the output is:

1) the recommended move is dependent upon search depth
and
2) non-determinism is surely a factor. Running a 2nd test under the exact same conditions will likely produce different results.

Also, running multiPV = 4 just slows and skews the results, as it allocates too much PC search time to non-recommended moves. One gets faster and better output with PV=1.

All the very best,
-Steve-