Big boost in strength for LC0 by simply changing VGA card

[brianr]

...I am doing another match now, but this time using a Silvesuite book and things are not going as well. At the moment after 60 games Leela is down by 2 games with score at +11, =36,-13. I looked at some of the openings from book and many openings had SF up by over +1 from the opening and Leela on many openings was very pessimistic about the position it had to start with. Most likely I will need a more neutral book. I wanted to try out the Nunn short book but couldn't find it somehow.

Books are actually a handicap for Leela. The Silver and Nunn books are great, just not as helpful with Leela as with "traditional" alpha/beta engines. Suggest using only a two ply (1 move each side) only book depth.

[ankan]

Card I am using is now an RTX 2080Ti which reading on the forums seems to indicate would do very well. The previous card was a Quadro M4000 with around 1700 Cuda cores while the RTX 2080Ti has about 3.5x that.

I would expect a bigger jump in NPS going from Quadro M4000 to the RTX 2080Ti (something like 10x or even more).
Make sure you have --backend=cudnn-fp16 as command line option (or select it from GUI's engine settings). lc0 utilizes tensor cores only with the cudnn-fp16 backend.
Increasing the NNCache size (from the default 200k to something like 10 million) can also boost NPS a bit without any side-effects (add --nncache=10000000 to command line settings or in the GUI's engine settings).

[M ANSARI]

Card I am using is now an RTX 2080Ti which reading on the forums seems to indicate would do very well. The previous card was a Quadro M4000 with around 1700 Cuda cores while the RTX 2080Ti has about 3.5x that.

I would expect a bigger jump in NPS going from Quadro M4000 to the RTX 2080Ti (something like 10x or even more).
Make sure you have --backend=cudnn-fp16 as command line option (or select it from GUI's engine settings). lc0 utilizes tensor cores only with the cudnn-fp16 backend.
Increasing the NNCache size (from the default 200k to something like 10 million) can also boost NPS a bit without any side-effects (add --nncache=10000000 to command line settings or in the GUI's engine settings).

Thanks for this advice. I did not have those settings so I am guessing this will improve things. I now get 52kn/s on the start position (not sure what it was before the setting change) using the network 32591. I am guessing there is probably some better networks out there ... but really everything is so much in flux right now that I feel things will be changing a lot for while. I was reading up on Tensor cores and it makes for some interesting reading. Apparently this is all done for AI with Google having been the starter of this but several other companies are starting to make even bigger headway in that field (nvidia being one of them). My guess is these Tensor cores are similar what the Pentium Pro a little over 2 decades ago was then. The hardware and software to use the hardware will probably see an exponential improvement as it becomes more mainstream. If LC0 is so strong already, I can't imagine what things will be like in a few years. There are some pretty huge holes in LC0 evaluation ... in one game I saw that it just did not realize how strong it was for the other side to gain 2 queens ...somehow it just gave it almost 1 pawn advantage for the extra queen while SF was at +10 and obviously SF was correct. These things will be patched up soon I think. I remember not too long ago Rybka in its prime, on my 8 core 5 Ghz system was crushing everything on Playchess ... but then in one game it completely missed a wrong bishop endgame and went head first into it thinking it was winning while it was a draw. Those type of things need to be patched up and if the rate of progress continues, it will only be a matter of time.

[jmartus]

Net 32335 is the strongest one. Should try it with that one.

[carldaman]

Net 32335 is the strongest one. Should try it with that one.

If that's the case, why did they pick the later 32742 for TCEC?

It is very hard to get non-contradictory information about LCO these days.

[jmartus]

I'm not 100 percent sure i was just going off someones ccrl rating list in the discord forum.

[Nay Lin Tun]

I'm not 100 percent sure i was just going off someones ccrl rating list in the discord forum.

ccrl rating list by MGTO is pretty incorrect indicator of NN rating for LTC or strong hardware like TCEC/CCcc. Lco is practically a hybrid engine and the strength of engine is the result of a combination of NN's policy strength(similar to static evaluation of Stockfish at 0 node) and value head strength( MCTS search and its evaluation which is similar to Stockfish evaluation at depth 30~50 etc ).
When bullet with 4k nps like ccrl list, the value of policy head matters cos the search is very limited.
But it Tcec/cccc, they have 60~ 100knps and also VLTC, and the better value head matters more. According to test results of big hardwares, 32700+ or 32800+ outperform the network u mentioned above.

[yanquis1972]

suspect there's been a significant gain in conjunction with the drop in self-elo, which seems to've recovered. a lot of complex endgames seem soft-solved; winning scores (~75%/~+2.50) where SF might give half/three quarters pawn advantage with fairly deep search.

this 1min+1s bullet game on weak hardware (~2.5MN/s SF10; ~6kn/s Lc0) deserves some much deeper analysis but is a complete spectacle without it -- https://lichess.org/fAEh9NgM



[pgn][Event "Sf10 vs lc0 id820vs876vs896"] [Site "PC"] [Date "2019.01.26"] [Round "1"] [White "Stockfish 10 1cpu"] [Black "Lc0 id32909"] [Result "0-1"] [BlackElo "2980"] [ECO "C77"] [Opening "Spanish"] [Time "19:01:27"] [Variation "Anderssen Variation"] [WhiteElo "2880"] [TimeControl "60+1"] [Termination "adjudication"] [PlyCount "97"] [WhiteType "program"] [BlackType "program"] 1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 (Ba4 Nf6 Bxc6 dxc6 d3 Bd6 O-O Qe7 h3 O-O b3 Be6 Bb2 Nd7 Nbd2 f6 a4 Qe8 Re1 Bc5 d4 Bd6 Nc4 Bxc4 bxc4) {+0.63/22 5} Nf6 (Nf6 O-O Be7 Re1 b5 Bb3 d6 c3 O-O h3 Bb7 d4 Re8 Nbd2 Bf8 a4 h6 Bc2 Rb8 Bb1 Bc8 Nf1 b4 Bd2 bxc3 Bxc3 exd4 Nxd4 Ne5 Nf3 Bb7 Nxe5 dxe5) {-0.55/12 1} 5. d3 (d3 b5 Bb3 Be7 a4 b4 a5 O-O O-O d6 Nbd2 h6 h3 Be6 Bxe6 fxe6 b3 Nd7 Bb2 Qe8 Re1 Qg6 Kh1) {+0.68/21 1} Bc5 (Bc5 O-O b5 Bb3 d6 c3 h6 Re1 O-O h3 Bb6 Nbd2 Ne7 Nh4 Be6 Nf1 Ng6 Nxg6 Bxb3 Qxb3 Re8 Nh4) {-0.30/10 1} 6. Bxc6 (Bxc6 dxc6 O-O Qd6 b3 Be6 Bb2 Nd7 Ng5 O-O-O Nd2 Rhe8 a4 Kb8 Qh5 g6 Qh6 Rh8 Ndf3 f6 Nxe6 Qxe6 Qd2) {+0.77/20 0} dxc6 (dxc6 O-O Qe7 Nbd2 Bg4 Nc4 Nd7 h3 Bh5 Bd2 f6 Rb1 Qf7 b4 Ba7 a4 O-O Ne3 Rfd8) {+0.04/9 1} 7. O-O (O-O Qd6 b3 Be6 Bb2 Nd7 Ng5 O-O-O Nd2 Kb8 a4 Rhe8 Qe2 f6 Nxe6 Qxe6 Rfe1 Bd4 Bxd4 exd4 Qh5 h6 f4 Qe7 f5) {+0.67/21 1} Qd6 (Qd6 Nbd2 Be6 Qe2 h6 a3 a5 b3 Nd7 Nc4 Bxc4 dxc4 Qe6 Bb2 O-O Rfd1) {+0.01/10 4} 8. Bd2 (Bd2 Be6 a4 Nd7 Ng5 O-O Nc3 h6 Nxe6 Qxe6 Qf3 Bb4 Qg3 Qg6 Qh3 Nc5 f4 exf4 Rxf4 Ne6 Rg4 Qf6 Qxh6 Qxh6 Bxh6 f6) {+0.57/23 5} Be6 (Be6 Bc3 Nd7 Ng5 h5 Nxe6 Qxe6 Nd2 O-O-O b4 Ba7 a4 g5 b5 g4 bxa6 bxa6 Qe2 h4 Kh1 Rdg8) {+0.11/9 2} 9. Bc3 (Bc3 Nd7 Ng5 O-O Nd2 Rae8 a3 b5 Nxe6 Rxe6 h3 Bd4 Bxd4 Qxd4 b3 Qd6 a4 Rd8 g3 Qe7 Kg2 h6 h4 c5) {+0.62/22 1} Ng4 (Ng4 Nbd2 f6 Qe2 O-O-O Nc4 Qd7 b4 Ba7 a4 h5 Rab1 h4 h3 Nh6 b5 cxb5 axb5 axb5) {+0.12/9 1} 10. h3 (h3 h5 Nbd2 O-O-O Ng5 Ne3 Nxf7 Bxf7 fxe3 Bxe3 Kh2 Be6 Nf3 Bd4 Ng5 Kb8 Qd2 Rdf8 Bxd4 Rxf1 Rxf1 exd4 Qf4 Qxf4 Rxf4 Bxa2 b3 Ka7) {+0.72/21 2} h5 (h5 Nbd2 f6 Nb3 Bb6 Nfd2 O-O-O Qe2 Nh6 Nc4 Bxc4 dxc4 Nf7 Rad1 Qe7 a3 Ng5 Rxd8) {+0.54/9 1} 11. Nbd2 (Nbd2 O-O-O Ng5 Ne3 Nxf7 Bxf7 fxe3 Bxe3 Kh2 Bf4 g3 Bxd2 Qxd2 Be6 Qe3 Kb8 Qg5 Qd7 Qxe5 Bxh3 Qxg7 Qc8 Rf4 h4 Rxh4 Rhg8 Qh6 Be6 Qe3 Rdf8 Bd4 Rg6) {+0.86/20 0} f6 (f6 Nb3 Bb6 Qe2 O-O-O Rad1 Qe7 d4 Qf7 Rfe1 Bc4 Qd2 g5 Qc1 Nh6 dxe5) {+0.56/10 1} 12. Nb3 (Nb3 Ba7 Qe2 O-O-O d4 Qe7 Rad1 g5 a4 Kb8 Na5 Bc8 Nc4 exd4 Nxd4 Nh6 Na5 Rd6 Nc4) {0.00/24 10} Bb6 (Bb6 Qe2 O-O-O Rad1 Qe7 a4 Kb8 a5 Ba7 d4 Qf7 dxe5 Bc4 Rxd8 Rxd8 Qe1 Bxf1 Qxf1) {+0.79/10 2} 13. Qe2 (Qe2 O-O-O Rad1 Kb8 Rd2 g5 d4 Qe7 a4 Bxb3 cxb3 exd4 Nxd4 Bxd4 Bxd4 Ne5 Rfd1 c5 Bxe5 Rxd2 Qxd2 Qxe5) {-0.05/21 4} O-O-O (O-O-O Rad1 Qe7 a4 Kb8 d4 Qf7 Na5 Bxa5 Bxa5 Bc4 Rd3 exd4 b3) {+1.00/9 1} 14. Nfd2 (Nfd2 Qe7 Nc4 Ba7 Nca5 Bb6) {0.00/23 2} Qe7 (Qe7 Nc4 Ba7 a4 g5 Bd2 Rdf8 Qe1 Rfg8 Bb4 c5 Ba3 Nh6 Qe3 b6 a5 b5) {+0.99/9 4} 15. Nc4 (Nc4 Ba7 Nca5 Bb6) {0.00/22 0} Ba7 (Ba7 a4 g5 Bd2 Rdf8 Rae1 Kb8 Nba5 Nh6 Be3 Bxe3 Qxe3 Nf7 b4 Bxc4 Nxc4 Qxb4 Rb1) {+0.98/10 0} 16. Nba5 (Nba5 Nh6 Bd2 Nf7 b4 Qd7 Be3 Bxe3 Nxe3 Ng5 h4 Nf7 Nb3 Bxb3 axb3 g5 Rfd1 Kb8 d4 exd4 Nf5 Qe6 Ra5 d3 Rxd3 Rxd3 Qxd3 gxh4 Nxh4) {+0.01/25 12} g5 (g5 Bd2 Rdf8 Rae1 Kb8 b4 Nh6 Be3 Bxe3 Qxe3 Nf7 a4 Nd8 Rb1 Bxc4 dxc4 b6 Nb3) {+0.71/9 3} 17. Nxb7 (Nxb7 Kxb7 Na5 Ka8 d4 exd4 Qxa6 Bc8 Qxc6 Kb8 Qb5 Ka8) {0.00/22 2} Kxb7 (Kxb7 Na5 Ka8 Nxc6 Qd6 Nxa7 Rd7 Rfd1 Kxa7 hxg4 hxg4 Qe3 Kb7 Kf1 Rh2 Ke2 Rxg2 Rg1 Rh2 Rh1 Rdh7 Rxh2) {+0.31/11 1} 18. Na5+ (Na5 Ka8 Nxc6 Qd7 Nxd8 Qxd8 Bd2 Nh6 Be3 Bxe3 Qxe3 Qd6 Rfb1 g4 h4 f5 b4 Kb7 b5 a5 a4 f4 Qd2 g3 fxg3 Ng4 gxf4 exf4 Qxa5) {+0.03/22 0} Ka8 (Ka8 Nxc6 Qd6 Nxa7 Rd7 Rfb1 Kxa7 hxg4 hxg4 Qe3 Kb7 Kf1 Rh2 Ke2 Rxg2 Rg1 Rh2 Rh1 Rdh7 Rxh2 Rxh2 Rg1) {+0.10/11 1} 19. Nxc6 (Nxc6 Qd7 Nxd8 Qxd8 Rad1 Kb7 Bd2 Qe7 a4 Nh6 Be3 Bb6 Bxb6 cxb6 a5 g4 h4 bxa5 d4 Nf7 Qe3 Qb4 b3 Rc8 d5 Bd7 c4 a4 bxa4 Rxc4) {-0.04/24 4} Qd6 (Qd6 Nxa7 Rd7 a4 Kxa7 hxg4 hxg4 Qe3 Kb7 Rfb1 Rdh7 Kf1 Rh2 Ke2 Rxg2 Rg1 Rgh2 Rg3 R8h3) {+0.02/10 3} 20. Nxd8 (Nxd8 Qxd8 Rad1 Qd6 Bd2 Kb7 b3 Nh6 Be3 g4 Bxa7 Kxa7 Qe3 Kb7 h4 c5 Rb1 f5 exf5 Nxf5 Qe4 Qd5 Rbe1 Nxh4 Qxd5 Bxd5 Rxe5 Bxg2 Rfe1 Bf3 Rxc5) {-0.03/23 4} Qxd8 (Qxd8 Bd2 Qd6 Rad1 Bc8 c3 f5 Bxg5 Qg6 h4 Rf8 exf5 Qxf5 d4) {+0.83/9 3} 21. Rad1 (Rad1 Qd6 Bd2 Kb7 b3 Qe7 c3 Bb6 c4 Nh6 b4 Nf7 Be3 Qe8 Rb1 g4 h4 Bd7 a3 f5 Qc2 fxe4 dxe4) {+0.07/20 0} Qd6 (Qd6 Bd2 Bc8 c3 Qe6 b4 Nh6 Be3 Bxe3 Qxe3 Nf7 b5 Qb6 bxa6) {+0.96/8 1} 22. Bd2 (Bd2 Nh6 Be3 h4 Bxa7 Kxa7 Qf3 Qe7 Qe3 Kb7 d4 exd4 Rxd4 Nf7 Ra4 Ne5 Rd1 g4 hxg4 Bxg4 Qb3 Ka7 Rd5 Be6 Qe3 Kb7 Qb3) {0.00/21 1} Bd7 (Bd7 c3 f5 Bxg5 Qg6 Bh4 f4 Kh1 Rg8 Rg1 Bb5 Rd2 Kb7 b4) {+1.04/8 2} 23. b3 (b3 Kb7 Qe1 Nh6 Be3 Bb6 Bxb6 Qxb6 Qe3 Nf7 Qxb6 Kxb6 f3 Nd6 d4 exd4 Rxd4 Re8 Re1 g4 hxg4 hxg4 Kf2 Rg8 Red1 gxf3) {+0.03/22 4} f5 (f5 Bxg5 f4 c3 Qg6 h4 Nf6 d4 Nxe4 Rfe1 Bc6 Qxa6 Rg8 f3) {+1.36/9 1} 24. Bxg5 (Bxg5 Qg6 Bh4 Nh6 Kh1 f4 Rg1 Nf7 Qd2 Qd6 Qe2) {0.00/22 1} f4 (f4 Kh1 Bd4 Bh4 Rg8 b4 Qg6 Rg1 Qc6 c3 Bb6 d4 Qxc3 Qxa6) {+1.11/9 2} 25. Bh4 (Bh4 Nh6 c3 Qg6 Kh1 Nf7 a4 Rg8 Rg1 Nd6 b4 Kb7 Rb1 Be6 Ra1 Ka8 Rad1 Bb6 Rb1 Bd7) {+0.31/20 1} Bd4 (Bd4 Kh1 Rg8 Rc1 Bc3 Rb1 Kb7 b4 Qe6 Rb3 Bd4 c3 Bb6 c4 Bd4 b5) {+0.94/8 2} 26. Rc1 (Rc1 Bb6 c3 Nh6 d4 Rg8 Kh1 Qg6 Rg1 exd4 cxd4 Bxd4 Rxc7 Bxh3 Qf3 Bc8 Qxf4 Qg4 Qxg4) {+0.49/20 1} Bc3 (Bc3 Kh1 Qe6 Qf3 Rf8 Rb1 Kb7 Rfd1 Rg8 a4 c5 b4 cxb4) {+1.14/8 1} 27. hxg4 (hxg4 hxg4 g3 Kb7 a4 Qg6 Rcd1 Qb6 d4 exd4 Qc4 Bc6 Bf6 Rh3 Qe6 fxg3 Rd3 gxf2 Rxf2 Re3 Rxe3 dxe3 Re2 Bxf6 Qxf6 Bxe4 Qxb6 cxb6 Rxe3) {+1.01/20 1} hxg4 (hxg4 g3 Qg6 Rfd1 Kb7 Rb1 Qh6 d4 exd4 Qc4 Bc6 b4 fxg3 fxg3) {+1.94/8 1} 28. g3 (g3 Qb6 d4 exd4 Qc4 fxg3 fxg3 Bb5 Qd5 Bc6 Qc4) {0.00/24 1} Kb7 (Kb7 Kh2 Rh5 Rh1 Qh6 Qf1 Bd4 Kg2 f3 Kg1 Bxf2 Kxf2 Qd2 Kg1 Qe3 Kh2 Qd2 Kg1 Qe3 Kh2 Qd2 Kg1) {+1.93/8 2} 29. Rfd1 (Rfd1 Qb6 Qf1 Rxh4 gxh4 g3 d4 Bxd4 Rxd4 Qxd4 c4 gxf2 Qxf2 Qxe4 Qd2 Qf5 Kf2 Qh3 Ke1 Qh1 Ke2 Qe4 Kf2 Bc6 Rg1 Qf3 Ke1 Qe4) {0.00/22 1} Rh5 (Rh5 a4 Qh6 d4 exd4 Rb1 f3 Qc4 Rxh4 gxh4 Qxh4 Qd5 Bc6 Qe5 Bb4 Qg3 Qh8) {+2.74/8 2} 30. Qf1 (Qf1 Qh6 Qg2 Bb2 Rb1 Bd4 c3 Bxc3 Rbc1 Bd4 Rc2 Bb6 a4 Qd6 Kf1 Qh6 Rc3 Bd4 Rc4 Bb6 a5 Bxa5 Qg1 Bb6 b4) {+0.28/21 0} Qh6 (Qh6 Qg2 f3 Qh2 Bd2 Ra1 Bg5 d4 Bxh4 gxh4 Rxh4 Qxh4 Qxh4 dxe5 g3) {+7.30/8 1} 31. Qg2 (Qg2 f3 Qh1 Bd2 Rxd2 Qxd2 Rf1 Qxc2 Kh2 Qxa2 Qg1 Bb5 Rc1 Rh6 Qe1 Qe2 Kg1 Rc6 Bg5 Rxc1 Qxc1 Qxd3 Be3 Qxb3 Bd2 Qd3 Ba5 c5 Qe1 Qc2 Bc3) {-2.30/27 1} f3 (f3 Qh2 Bd2 Rb1 Bg5 d4 Bxh4 gxh4 Rxh4 Qxe5 Rh1) {+12.97/9 1} 32. Qh1 (Qh1 Bd2 Rxd2 Qxd2 Rf1 Qxc2 Kh2 Qxa2 Qg1 Bb5 Rc1 Qxb3 Qd1 Qxd3 Qxd3 Bxd3 Kg1 Bxe4 Rc4 Bd3 Rxg4 a5 Rg7 Rh6 Re7 e4 Bg5 Rb6 Re5 Rb1 Kh2 Rb5 Rxb5 Bxb5 g4) {-2.21/26 0} Bd2 (Bd2 Rb1 Bg5 d4 Bxh4 gxh4 Rxh4 Qxh4 Qxh4 Rbc1 g3 fxg3) {+18.50/8 1} 33. Rxd2 (Rxd2 Qxd2 Rf1 Qxc2 Kh2 Qxa2 Qg1 Bb5 Rc1 Qxb3 Qd1 Qxd3 Qxd3 Bxd3 Kg1 Bxe4 Rc4 Bd3 Rxg4 a5 Bd8 e4 Rg5 Rh8 Be7 Ra8 Re5 Ra6 Ba3 a4 g4 Rb6 Kh2 Rb3 Bc1 a3 Ra5) {-2.29/29 1} Qxd2 (Qxd2 Rf1 Qxc2 Bg5 Rxg5 Qh7 Qc6 Qe7 Rh5 Qb4 Qb6) {+23.23/7 2} 34. Ra1 (Ra1 Qxc2 Qh2 Qb2 Rd1 Qxa2 Kh1 Bb5 Qg1 Qxb3 Qe1 Bxd3 Rc1 Rh7 Kg1 Rd7 Bg5 Bb5 Be3 Qa4 Bd2 Qd4 Be3 Qxe4 Qc3 Qd5 Qc2 e4 Qc5 Qxc5 Rxc5 c6 Re5 Rd1 Kh2) {-2.28/29 2} Qxc2 (Qxc2 Kh2 Qxf2 Qg2 Qxg2) {+24.77/7 2} 35. Qh2 (Qh2 Qb2 Rd1 Qxa2 Kh1 Bb5 Qg1 Rh6 Qe1 Qxb3 Kg1 Bxd3 Rc1 Qb5 Qc3 Rc6 Qxc6 Qxc6 Rxc6 Kxc6 Bf6 Kd6 Bg7 a5 Bf8 Kc6 Be7 a4 Ba3 Bxe4 Kf1 Bd3 Ke1 e4 Kd2) {-2.33/27 1} Qb2 (Qb2 Rd1 Qe2 Ra1 Qxd3 Kh1 Qxe4 Qg1 Bb5 Rc1 Qd4 Rd1) {+31.54/6 3} 36. Re1 (Re1 Qxa2 Kh1 Qd2 Ra1 Rh6 Qg1 Bb5 Qe1 Qxd3 Kg1 Qxb3 Rc1 Rd6 Be7 Rd4 Bc5 Ra4 Rb1 Qd3 Bb4 Ra2 Rd1 Qe2 Bc3 Ra4 Qd2 Qxd2 Rxd2 Rxe4 Rd5 Rc4 Bxe5 Rc1 Kh2 Rc2) {-2.84/29 4} Qd2 (Qd2 Ra1 Qc3 Rd1 Qc2 Ra1 Qxd3 Qh1 Qxe4 Kh2 Qe2 Kg1) {+35.65/6 2} 37. Ra1 (Ra1 Qb2 Re1 Qxa2 Kh1 Rh6 Qg1 Bb5 Rc1 Qe2 Kh2 Bxd3 Qe1 Rc6 Kg1 Qxe1 Rxe1 Rb6 Rd1 Rxb3 Bf6 c5 Bxe5 c4 Bf4 Kc6 Kh2 Rb5 Kg1 a5 e5 Kd5 e6 Kxe6) {-2.64/28 1} Qc3 (Qc3 Rd1 Qc2 Ra1 Qxd3 Kh1 Qxe4 Qg1 Bb5 Qc1 Qd4 Kg1) {+29.40/6 1} 38. Rd1 (Rd1 Qc2 Ra1 Qb2 Rd1 Qxa2 Kh1 Qc2 Ra1 Bb5 Qg1 Rh6 Qf1 Bxd3 Qe1 Qe2 Kg1 Qxe1 Rxe1 Rb6 Rd1 Rxb3 Bf6 c5 Bxe5 Bxe4 Rc1 Kc6 Bd4 Rb5 Bc3 a5 Ra1) {-2.66/29 1} Qc2 (Qc2 Ra1 Qb2 Rd1 Qe2 Ra1 Qxd3 Kh1 Qxe4 Qg1 Bb5 Rc1) {+29.61/7 1} 39. Ra1 (Ra1) {-2.71/31 4} Qb2 (Qb2 Rd1 Qe2 Ra1 Qxd3 Kh1 Qxe4 Qg1 Bb5 Qc1 Qd4 Kg1) {+28.14/7 2} 40. Rd1 (Rd1 Qxa2 Kh1 Qc2 Ra1 Qxd3 Qg1 Qxe4 Qe1 Qe2 Qb4 Bb5 Kg1 Rxh4 gxh4 g3 Qe1 Qc2 Qe3 gxf2 Qxf2 Qe4 Re1 Qg4 Kh2 Qf4 Kh1 Be2 Rxe2 fxe2 Qxe2 Qxh4 Kg2 Qf4 Qe1 Qd4 Kg3 Kb6 Kf3 e4 Ke2) {-3.02/30 2} Qe2 (Qe2 Ra1 Bb5 Kh1 Bxd3 Qg1 Qxe4 Rc1 Qd4 Qe1 e4) {+27.86/6 2} 41. Ra1 (Ra1 Bb5 Kh1 Rh6 Qg1 Qb2 Rc1 Qxa2 Qe1 Qe2 Kg1 Rc6 Bg5 Bxd3 Be3 Qxe1 Rxe1 Rc3 b4 Rc4 Bd2 Rc2 Bg5 Re2 Rc1 Bxe4) {-2.12/22 0} Bb5 (Bb5 Qh3 gxh3 Bg5 Kb6 Bd2 h2 Kh1) {+20.99/6 2} 42. Kh1 (Kh1 Rh6 Qg1 Qb2 Rc1 Bxd3 Rd1 Be2 Rc1 Qxa2 Qe1 Bb5 Kg1 Qxb3 Bg5 Rd6 Qe3 Qe6 Qc5 Rc6 Qd5 Qxd5 exd5 Rd6 Rc5 Rd7 Bf6 e4 Be5) {-2.00/22 0} Bxd3 (Bxd3 Qg1 Qxe4 Rc1 Qe2 Qe1 Qxe1 Rxe1 e4 Kg1 Ra5 Ra1 Kc6) {+20.43/6 2} 43. Qg1 (Qg1 Qxe4 Qe1 Qxe1 Rxe1 e4 Kg1 Rc5 Bf6 Rc2 a3 c5 Bg5 Rb2 Be3 Rxb3 Bxc5 Rc3 Bd4 Rxa3 Rc1 a5 Ra1 Rxa1 Bxa1 Kc6 Bb2 Kc5 Ba3 Kd5 Bc1) {-1.91/23 0} Qxe4 (Qxe4 Rc1 Qe2 Qe1 Qxe1 Rxe1 e4 Kg1 Ra5 a4 Rc5 Bf6 Rc2 Bg5 Rb2) {+20.97/7 1} 44. Qe1 (Qe1 Qxe1 Rxe1 e4 Kg1 Rc5 Bf6 Rc2 a4 Kb6 b4 Rc4 Be7 Kc6 a5 Kb5 Bc5 Rc3 Ra1 Rc2 Re1 Rc4 Kh2 Rc3 Ra1 Rc2 Kg1 Rb2) {-1.87/25 0} Qxe1+ (Qxe1 Rxe1 e4 Kg1 Rc5 Bf6 Rc2 Bg5 Rxa2 b4 Rb2) {+28.22/7 1} 45. Rxe1 (Rxe1 e4 Kg1 Rc5 Bf6 Rc2 Bd4 Rxa2 Rc1 Bc2 b4 Bd3 Kh2 Ra3 Be3 Rb3 Bf4 Rb2 Kg1 c6 Rc5 Rxb4 Rg5 Rb5 Rxg4 a5 Rg7 Kb6 Be3 c5 Rg6 Kb7 Rg7 Kc6 g4 Kd5 Kh2) {-2.48/26 1} e4 (e4 Kg1 Rc5 Bf6 Rc2 a3 c5 Bg5 Ra2 Rc1 Rxa3 Rxc5) {+26.72/7 2} 46. Kg1 (Kg1 Rc5) {-2.75/28 2} Rc5 (Rc5 Bf6 Rc2 a3 c5 Bg5 Ra2 Rc1 Rxa3 Rxc5 Rxb3 Be3 Bb5) {+23.45/6 2} 47. Bf6 (Bf6 Rc2 a4 c5 Bg5 Ra2 Rc1 c4 bxc4 Rxa4 Rc3 Kc6 Rb3 Kc5 Rb7 Ra1 Kh2 Kxc4 Rc7 Kb3 Rb7 Kc2 Bf6 Ra2 Bd4 a5 Rg7 Kb3 Rxg4 Kc4 Be3 Re2 Rg8 Rxe3 Rc8 Kb3 fxe3 f2) {-3.07/27 1} Rc2 (Rc2 a3 c5 Bg5 Ra2 Rc1 Rxa3 Rxc5 Rxb3 Be3 Rb5 Rg5 Rxg5 Bc1) {+25.20/6 1} 48. Bd4 (Bd4) {-3.10/25 0} Rxa2 (Rxa2 Rc1 Ra3 Rc3 a5 Kh2 Ra1 Rxc7 Kxc7 Bxa1 e3 Kg1) {+29.35/6 1} 49. Rc1 (Rc1 Bc2 Kf1 a5 Ra1 Rxa1 Bxa1 Kc6 Bb2 Kb5 Ba3 Bxb3 Ke1 Bd5 Bc1 Kc4 Be3 a4 Kd1 Kb3 Kc1 a3 Kb1 Kc3 Bc1 Kd3 Be3 Ke2 Bd4) {-3.21/22 0 Arena Adjudication} 0-1[/pgn]

[M ANSARI]

I did another 3_2 game, this time 50 games with the new settings and it seems score is very similar to the initial 20 game set. Lc0 is at +16 -6 and the rest draws after 40 games. This is around +90 ELO which is quite impressive. Again this is with no books. I have to say I am really impressed. There was one game where Lc0 crazily gave up its queen and went into an ending thinking it was equal. I mean I guarantee that 100% of all super GM's and every engine would disagree as it seemed there was no fortress in the endgame. SF had the evaluation at around +2.4 ... but it turns out that Lc0 knew that it could force off one pair of rooks and then get a fortress. I was following that game live and I thought that either Lc0 was a genius and or was going to go down in flames ... I guess after about 40 moves SF seemed to agree there was no progress so Lc0 was probably correct. For me that was mind boggling as the weakness of engines was that they could not understand fortresses with inferior material or had a blindness for the horizon effect. Not anymore I guess! Once the game is over I will try to filter that game out as I thought it was game changing.

[Javier Ros]

I did another 3_2 game, this time 50 games with the new settings and it seems score is very similar to the initial 20 game set. Lc0 is at +16 -6 and the rest draws after 40 games. This is around +90 ELO which is quite impressive. Again this is with no books. I have to say I am really impressed. There was one game where Lc0 crazily gave up its queen and went into an ending thinking it was equal. I mean I guarantee that 100% of all super GM's and every engine would disagree as it seemed there was no fortress in the endgame. SF had the evaluation at around +2.4 ... but it turns out that Lc0 knew that it could force off one pair of rooks and then get a fortress. I was following that game live and I thought that either Lc0 was a genius and or was going to go down in flames ... I guess after about 40 moves SF seemed to agree there was no progress so Lc0 was probably correct. For me that was mind boggling as the weakness of engines was that they could not understand fortresses with inferior material or had a blindness for the horizon effect. Not anymore I guess! Once the game is over I will try to filter that game out as I thought it was game changing.

I had the same impressions when I played the first games with Lc0. I am still amazed when I see how it is able to turn a game where Stockfish evaluates with advantage sacrificing a pair of pawns. The following game was played between SF10 on 6 cores of i7 980X and AntiFish125 on 2x1060 played at 5 minutes plus 1 second.

After eating the second pawn SF10 evaluates as +0.86



but the advanced black pawns of d and e worsen the coordination of the white pieces and the threat of the column g gives black advantage. After 29.g3 +0.50 if follows



[pgn][Event "BalsaAFx2_5p1"] [Site "J-PC"] [Date "2019.01.26"] [Round "9"] [White "Stockfish_10_x64_popcnt"] [Black "AF125x2"] [Result "0-1"] [ECO "B33"] [WhiteElo "2200"] [BlackElo "2200"] [PlyCount "119"] [EventDate "2019.??.??"] [TimeControl "300+1"] 1. e4 c5 2. Nf3 e6 3. d4 cxd4 4. Nxd4 Nc6 5. Nc3 Nf6 6. Ndb5 d6 7. Bf4 e5 8. Bg5 a6 9. Na3 b5 10. Bxf6 {0.44/25 15} gxf6 {-0.38/11 9} 11. Nd5 {0.57/24 3} Bg7 {-0.35/11 4} 12. Bd3 {0.87/23 2} Ne7 {-0.43/12 13} 13. Nxe7 {0.70/25 7} Qxe7 {-0.45/12 1} 14. c4 {0.72/25 11} f5 {-0.33/11 12} 15. O-O {0.82/25 7} O-O {-0.41/11 2} 16. Qh5 {0.46/25 8} h6 {-0.32/10 10} 17. Qe2 {0.55/27 17} Bb7 { -0.09/11 8} 18. Rad1 {0.66/24 1} Rad8 {-0.02/12 4} 19. exf5 {0.84/24 2} e4 { 0.13/12 8} 20. cxb5 {0.86/25 4} axb5 {0.11/12 4} 21. Bxb5 {0.86/24 3} d5 { 0.23/11 6} 22. f3 {0.65/25 7} e3 {0.29/11 5} 23. Nc2 {0.92/24 2} d4 {0.75/9 10} 24. a4 {0.66/26 15} Kh8 {1.00/10 10} 25. a5 {0.09/29 20} Rg8 {1.33/12 9} 26. a6 {0.00/24 4} Ba8 {1.36/11 5} 27. Ne1 {0.00/27 6} Qh4 {1.36/11 7} 28. Nd3 { -0.37/25 6} Rd5 {1.27/12 10} 29. g3 {0.50/25 2} Bf6 {1.32/12 10} 30. Be8 { -2.26/28 35} Rxf5 {2.59/12 7} 31. Bd7 {-2.36/27 7} Be5 {2.42/13 5} 32. Nxe5 { -2.71/27 6} Rxe5 {2.84/12 6} 33. Bg4 {-2.66/27 8} Qd8 {3.21/13 8} 34. Bh3 { -2.55/26 2} Qb6 {3.33/13 6} 35. Rfe1 {-2.69/30 18} Rd8 {3.55/14 6} 36. Rd3 { -3.45/29 33} Bc6 {4.26/10 12} 37. Bf1 {-3.47/25 8} Bb5 {5.53/10 9} 38. f4 { -3.83/25 7} Red5 {5.37/11 5} 39. Qd1 {-3.44/26 8} Kg8 {5.21/10 18} 40. Qg4+ { -3.99/26 21} Kf8 {5.63/10 5} 41. a7 {-4.29/28 14} Bxd3 {8.20/9 6} 42. Bxd3 { -4.45/22 0} Qxa7 {8.46/9 4} 43. Rf1 {-3.53/21 3} Qe7 {9.55/8 11} 44. f5 { -3.20/21 0} R5d6 {11.80/8 9} 45. b4 {-3.58/21 1} e2 {11.91/9 4} 46. Bxe2 { -3.25/24 2} d3 {11.30/9 3} 47. Bd1 {-4.00/22 1} Rd4 {11.92/9 6} 48. Rf4 { -3.58/23 0} Qe3+ {18.28/10 5} 49. Kg2 {-3.66/26 2} Rxf4 {20.53/10 3} 50. Qxf4 { -4.01/30 1} Qxf4 {24.37/10 4} 51. gxf4 {-4.83/30 3} Rd4 {23.50/9 11} 52. Kf3 { -4.46/25 0} Rxb4 {22.67/9 0} 53. Ke3 {-5.19/27 2} Ke7 {25.37/8 4} 54. Bf3 { -5.67/27 2} d2 {26.23/8 4} 55. Kxd2 {-5.83/28 1} Rxf4 {47.15/7 2} 56. Ke3 { -6.26/29 0} Rxf5 {48.54/7 3} 57. Kf2 {-5.47/25 0} Rxf3+ {128.00/3 4} 58. Kxf3 { -66.86/32 0} h5 {128.00/2 1} 59. Ke3 {-73.45/33 1} Kf6 {16.22/7 2} 60. Kf4 { adjudication -153.08/32 1 Arena Adjudication} 0-1 [/pgn]