Test position: Lc0 vs Stockfish, TCEC 18, game 65

peter · Post by **peter** » Thu Jul 02, 2020 12:39 am

zullil wrote: ↑Thu Jul 02, 2020 12:29 am Komodo-14 with Use LMR set to false. Takes more than 12 minutes to find Qxb4. (And much longer to show a positive score.)

info depth 27 time 746743 nodes 23483184798 score cp -1 lowerbound nps 31447478 hashfull 1000 tbhits 0 pv c4b4 e7b4 a3b4
info currmove c4b4 currmovenumber 1
info nodes 23483399820
info nodes 23548302286
info nodes 23614472881
info nodes 23680567129
info depth 27 time 753076 nodes 23691456733 score cp 0 nps 31459540 hashfull 1000 tbhits 0 pv c4b4 e7b4 a3b4 f7f5 e4f6 g8h8 d1d4 e8e7 e1d1 c8c7 f3g5 d8c8 g5e6 e7e6 f6d7 g6g5 h6g5 h8g7 d7c5 e6e8 g5f6 g7g6 d1d3 h7h6 d3g3 g6h7 d4h4 c7f7 h4h6 h7h6 g3h3 h6g6 h3g3 g6h6

Thanks for trying. So it could be a matter of LMRs also but probably not of LMRs only.

Ovyron · Post by **Ovyron** » Thu Jul 02, 2020 3:49 am

M ANSARI wrote: ↑Wed Jul 01, 2020 6:59 am This just shows that computer chess has reached a level of understanding that humans simply cannot follow. I consider myself quite a strong player and when I look at games between say Magnus Carlsen and Caruana ... I pretty much understand almost every move they make and can predict the majority of the moves. Doesn't mean I can play anywhere close to what they can, but I really can understand the logic and calculate the lines they do. But here, replaying the game ... I have absolutely no clue what is going on, and if they had "guess the next move" it really would be impossible to guess the majority of the moves. Of course I can replay the game and enjoy it, but OMG I just feel that the gap between human chess and engines has become gigantic!

You're not alone. In a video Hikaru Nakamura was analyzing a game between engines, and he admitted he had no idea what was going on (when a random rook move was clearly better than the alternatives and he just couldn't get why), and went as far as saying "this isn't chess" anymore.

The gap became gigantic since Leela, before people were still analyzing engine games and being able to explain the moves. Afterwards Leela is down a bunch of material and claiming big advantage while Stockfish is claiming 0.00 and all the humans know Leela is right as she goes to win the game, but they're not able to put why into words, into human-made concepts that don't apply to some positions.

To remain in the race Stockfish had to raise the bar and play moves that not even Leela on big hardware can choose or understand because they're waaay too deep, so humanity has to look at them with a telescope.

MikeB · Post by **MikeB** » Fri Jul 03, 2020 4:40 am

zullil wrote: ↑Thu Jul 02, 2020 12:29 am
peter wrote: ↑Wed Jul 01, 2020 10:24 pm Hi Louis!
zullil wrote: ↑Wed Jul 01, 2020 9:21 pm That's very fast! I haven't studied the code, but I'm guessing that much less LMR is being applied. Now that I'm curious, let me try Komodo with LMR totally disabled...
Schwabli (my chess- PC) is busy right now, so I'm curious for your results.
As for Michael Byrne's r- versions, there is presearch with different kind of pruning- techniques reduced, following ideas of Michael Nowak's HekaStockfish. One of the two Michaels should be best to tell more about it.

http://talkchess.com/forum3/viewtopic.p ... 73#p833973
http://talkchess.com/forum3/viewtopic.p ... 11#p837011
Komodo-14 with Use LMR set to false. Takes more than 12 minutes to find Qxb4. (And much longer to show a positive score.)

info depth 27 time 746743 nodes 23483184798 score cp -1 lowerbound nps 31447478 hashfull 1000 tbhits 0 pv c4b4 e7b4 a3b4
info currmove c4b4 currmovenumber 1
info nodes 23483399820
info nodes 23548302286
info nodes 23614472881
info nodes 23680567129
info depth 27 time 753076 nodes 23691456733 score cp 0 nps 31459540 hashfull 1000 tbhits 0 pv c4b4 e7b4 a3b4 f7f5 e4f6 g8h8 d1d4 e8e7 e1d1 c8c7 f3g5 d8c8 g5e6 e7e6 f6d7 g6g5 h6g5 h8g7 d7c5 e6e8 g5f6 g7g6 d1d3 h7h6 d3g3 g6h7 d4h4 c7f7 h4h6 h7h6 g3h3 h6g6 h3g3 g6h6

Pro Value is using multipv=8 and, in addition, is allocating more nodes to be searched at lower depth in certain positions , than it switches to a normal search after a bit. Horrible for micro bullet and bullet games - longer games and analysis seems ok. I was thinking about deactivating the code, but now I will reconsider.

peter · Post by **peter** » Fri Jul 03, 2020 7:42 am

MikeB wrote: ↑Fri Jul 03, 2020 4:40 am Pro Value is using multipv=8 and...

Thanks for clarifying, didn't know MultiPV to be involved too in presearch.

So if you chose "Tactical" in addition, is then the number of internal primaries of the Tactical- mode (if set higher then 3) presearched with less pruning or yet 8 only still?

MikeB · Post by **MikeB** » Fri Jul 03, 2020 11:30 pm

peter wrote: ↑Fri Jul 03, 2020 7:42 am
MikeB wrote: ↑Fri Jul 03, 2020 4:40 am Pro Value is using multipv=8 and...
Thanks for clarifying, didn't know MultiPV to be involved too in presearch.

So if you chose "Tactical" in addition, is then the number of internal primaries of the Tactical- mode (if set higher then 3) presearched with less pruning or yet 8 only still?

I have code so tactical , presearch and profound do not at play together all at the same time, since the results are undesirable
examples

Code: Select all

    if (mpv > 1)
          profound = false;

     if (!tactical && profound) yada yada

     if (tactical) {
         multiPV = pow(2, tactical);
        profound = false;
        }

tactical and mutlipv can play together, useful if you want to see best 4 moves let's say , but perform the search with multipv set to 8 - do display best 4 moves, but searching best 8 moves - set tactical to 3 ( 2^3=8 , 2 cubed) and multipv to 4.

Ovyron · Post by **Ovyron** » Sat Jul 04, 2020 1:32 am

And all of that is hidden because the engine only shows the info of the best move (so people could be using MultiPV=8 to do that and see the score of the other 7 moves, instead of looking only at the best move. )

peter · Post by **peter** » Sat Jul 04, 2020 8:08 am

MikeB wrote: ↑Fri Jul 03, 2020 11:30 pm I have code so tactical , presearch and profound do not at play together all at the same time, since the results are undesirable
examples
Code: Select all
    if (mpv > 1)
          profound = false;

     if (!tactical && profound) yada yada

     if (tactical) {
         multiPV = pow(2, tactical);
        profound = false;
        }
tactical and mutlipv can play together, useful if you want to see best 4 moves let's say , but perform the search with multipv set to 8 - do display best 4 moves, but searching best 8 moves - set tactical to 3 ( 2^3=8 , 2 cubed) and multipv to 4.

Thanks, so I guess I'm right to understand, that mpv switches off profound, but how about tactical and profound (yada yada?)?
Do I get it correctly too, that, if tactical chosen, that switches off profound fully too?

MikeB · Post by **MikeB** » Sat Jul 04, 2020 8:19 am

peter wrote: ↑Sat Jul 04, 2020 8:08 am
MikeB wrote: ↑Fri Jul 03, 2020 11:30 pm I have code so tactical , presearch and profound do not at play together all at the same time, since the results are undesirable
examples
Code: Select all
    if (mpv > 1)
          profound = false;

     if (!tactical && profound) yada yada

     if (tactical) {
         multiPV = pow(2, tactical);
        profound = false;
        }
tactical and mutlipv can play together, useful if you want to see best 4 moves let's say , but perform the search with multipv set to 8 - do display best 4 moves, but searching best 8 moves - set tactical to 3 ( 2^3=8 , 2 cubed) and multipv to 4.
Thanks, so I guess I'm right to understand, that mpv switches off profound, but how about tactical and profound (yada yada?)?
Do I get it correctly too, that, if tactical chosen, that switches off profound fully too?

Correct, enabling tactical, disables profound fully.,

M ANSARI · Post by **M ANSARI** » Sat Jul 04, 2020 10:24 am

Ovyron wrote: ↑Thu Jul 02, 2020 3:49 am
M ANSARI wrote: ↑Wed Jul 01, 2020 6:59 am This just shows that computer chess has reached a level of understanding that humans simply cannot follow. I consider myself quite a strong player and when I look at games between say Magnus Carlsen and Caruana ... I pretty much understand almost every move they make and can predict the majority of the moves. Doesn't mean I can play anywhere close to what they can, but I really can understand the logic and calculate the lines they do. But here, replaying the game ... I have absolutely no clue what is going on, and if they had "guess the next move" it really would be impossible to guess the majority of the moves. Of course I can replay the game and enjoy it, but OMG I just feel that the gap between human chess and engines has become gigantic!
You're not alone. In a video Hikaru Nakamura was analyzing a game between engines, and he admitted he had no idea what was going on (when a random rook move was clearly better than the alternatives and he just couldn't get why), and went as far as saying "this isn't chess" anymore.

The gap became gigantic since Leela, before people were still analyzing engine games and being able to explain the moves. Afterwards Leela is down a bunch of material and claiming big advantage while Stockfish is claiming 0.00 and all the humans know Leela is right as she goes to win the game, but they're not able to put why into words, into human-made concepts that don't apply to some positions.

To remain in the race Stockfish had to raise the bar and play moves that not even Leela on big hardware can choose or understand because they're waaay too deep, so humanity has to look at them with a telescope.

Actually before yesterday on Chessable game they had Harikrishna discussing this exact topic and how sometimes even deep analysis with Leela is hard as you simply cannot see why it sacrificed a pawn and it is just beyond human comprehension. Lc0 tends to sometimes give material for compensation that will only show up many moves later. Humans have a terrible time addressing compensation where the other side can play several moves but the forces are not coordinated just enough where you can get compensation by simply gaining one single tempo ... for a human that is beyond impossible to calculate, especially when there are 10's of other possible choices and counter choices. Of course one wrong move and that compensation will never be realized. I feel that the difference between GM's and strong players is timing a pawn sacrifices on d4 and e4 or even b4 breaks. Probably the king of these breaks is Magnus Carlsen and it is one of the things that makes him the strongest chess player on earth. Engines seem to take that type of play about 5 levels higher.

MMarco · Post by **MMarco** » Sun Aug 29, 2021 12:19 am

Stockfish 14 and Fat Titz got that one on depth 46 & 47 in these runs (16 Threads, 16 Gb):

[d]2rqr1k1/pp1bbp1p/4p1pB/4P3/1nQpN3/P4N2/5PP1/3RR1K1 w - - 0 1

Analysis by Stockfish 14:

Code: Select all

...
21.Qxd4 Nd5 22.Nd6 Bxd6 23.exd6 f6 24.Ng5 Nb6 25.Ne4 Kf7 26.Qb2 Rc4 27.Ng5+ Kg8 28.Ne4 
  The position is equal: = (0.00) (50% +4% =92% -4%)  Depth: 46/16   00:01:03  1152MN
21.Qxb4 
  The position is equal: = (0.08 ++) (51% +5% =92% -4%)  Depth: 47/40   00:01:34  1722MN
21.Qxb4 
  The position is equal: = (0.16 ++) (51% +6% =91% -3%)  Depth: 47/40   00:01:37  1785MN
21.Qxb4 
  White has an edge: = (0.29 ++) (53% +7% =90% -2%)  Depth: 47/40   00:01:43  1901MN
21.Qxb4 
  White is slightly better: +/= (0.47 ++) (54% +10% =88% -2%)  Depth: 47/40   00:01:48  1985MN
21.Qxb4 
  White is better: +/- (0.71 ++) (58% +16% =83% -1%)  Depth: 47/40   00:01:55  2118MN
21.Qxb4 
  White is better: +/- (1.04 ++) (64% +28% =72% -1%)  Depth: 47/40   00:02:02  2263MN
21.Qxb4 
  White is clearly better: +/- (1.47 ++) (74% +49% =51% -0%)  Depth: 47/40   00:02:13  2473MN
21.Qxb4 
  White is clearly better: +- (2.03 ++) (88% +76% =24% -0%)  Depth: 47/40   00:02:25  2701MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Re7 25.Red1 Rc7 26.Ng5 Qc8 27.b5 b6 28.Rd6 Rb7 29.R1d2 Rc7 30.Kh2 Rb7 31.R2d4 Rc7 32.Rd1 Rb7 33.R1d2 Qc4 34.Bf8 Be8 35.R2d4 Qxb5 36.f4 a5 37.Bxe7 Rxe7 38.Rd8 Qc6 39.R4d6 Qc4 40.Nh3 Qb5 41.Ra8 Kg7 42.Rdd8 Bc6 43.Rg8+ Kh6 
  White is clearly better: +/- (1.44) (74% +48% =52% -0%)  Depth: 47/55   00:02:28  2755MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Re7 25.Red1 Rc7 26.Ng5 Qc8 27.b5 b6 28.Rd6 Rb7 29.R1d2 Rc7 30.Kh2 Rb7 31.R2d4 Rc7 32.Rd1 Rb7 33.R1d2 Qc4 34.Bf8 Be8 35.R2d4 Qxb5 36.f4 a5 37.Bxe7 Rxe7 38.Rd8 Qc6 39.R4d6 Qc4 40.Nh3 Qb5 41.Ra8 Kg7 42.Rdd8 Bc6 43.Rg8+ Kh6 
  White is clearly better: +/- (1.52 ++) (76% +52% =48% -0%)  Depth: 48/21   00:03:05  3472MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Re7 25.Red1 Rc7 26.Ng5 Qc8 27.b5 b6 28.Rd6 Rb7 29.R1d2 Rc7 30.Kh2 Rb7 31.R2d4 Rc7 32.Rd1 Rb7 33.R1d2 Qc4 34.Bf8 Be8 35.R2d4 Qxb5 36.f4 a5 37.Bxe7 Rxe7 38.Rd8 Qc6 39.R4d6 Qc4 40.Nh3 Qb5 41.Ra8 Kg7 42.Rdd8 Bc6 43.Rg8+ Kh6 
  White is clearly better: +- (1.61 ++) (78% +56% =44% -0%)  Depth: 48/21   00:03:13  3622MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Re7 25.Red1 Rc7 26.Ng5 Qc8 27.b5 b6 28.Rd6 Rb7 29.R1d2 Rc7 30.Kh2 Rb7 31.R2d4 Rc7 32.Rd1 Rb7 33.R1d2 Qc4 34.Bf8 Be8 35.R2d4 Qxb5 36.f4 a5 37.Bxe7 Rxe7 38.Rd8 Qc6 39.R4d6 Qc4 40.Nh3 Qb5 41.Ra8 Kg7 42.Rdd8 Bc6 43.Rg8+ Kh6 
  White is clearly better: +- (1.73 ++) (81% +62% =38% -0%)  Depth: 48/24   00:03:22  3799MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Re7 25.Red1 Rc7 26.Ng5 Qc8 27.b5 b6 28.Rd6 Rb7 29.R1d2 Rc7 30.Kh2 Rb7 31.R2d4 Rc7 32.Rd1 Rb7 33.R1d2 Qc4 34.Bf8 Be8 35.R2d4 Qxb5 36.f4 a5 37.Bxe7 Rxe7 38.Rd8 Qc6 39.R4d6 Qc4 40.Nh3 Qb5 41.Ra8 Kg7 42.Rdd8 Bc6 43.Rg8+ Kh6 
  White is clearly better: +- (1.91 ++) (85% +71% =29% -0%)  Depth: 48/24   00:03:40  4146MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Re7 25.Red1 Rc7 26.Ng5 Qc8 27.b5 b6 28.Rd6 Rb7 29.R1d2 Rc7 30.Kh2 Rb7 31.R2d4 Rc7 32.Rd1 Rb7 33.R1d2 Qc4 34.Bf8 Be8 35.R2d4 Qxb5 36.f4 a5 37.Bxe7 Rxe7 38.Rd8 Qc6 39.R4d6 Qc4 40.Nh3 Qb5 41.Ra8 Kg7 42.Rdd8 Bc6 43.Rg8+ Kh6 
  White is clearly better: +- (2.15 ++) (90% +80% =20% -0%)  Depth: 48/24   00:03:59  4517MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Re7 25.Red1 Rc7 26.Ng5 Qc8 27.b5 b6 28.Rd6 Rb7 29.R1d2 Rc7 30.Kh2 Rb7 31.R2d4 Rc7 32.Rd1 Rb7 33.R1d2 Qc4 34.Bf8 Be8 35.R2d4 Qxb5 36.f4 a5 37.Bxe7 Rxe7 38.Rd8 Qc6 39.R4d6 Qc4 40.Nh3 Qb5 41.Ra8 Kg7 42.Rdd8 Bc6 43.Rg8+ Kh6 
  White is clearly better: +- (2.48 ++) (94% +89% =11% -0%)  Depth: 48/24   00:04:21  4948MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Re7 25.Red1 Rc7 26.Ng5 Qc8 27.b5 b6 28.Rd6 Rb7 29.R1d2 Rc7 30.Kh2 Rb7 31.R2d4 Rc7 32.Rd1 Rb7 33.R1d2 Qc4 34.Bf8 Be8 35.R2d4 Qxb5 36.f4 a5 37.Bxe7 Rxe7 38.Rd8 Qc6 39.R4d6 Qc4 40.Nh3 Qb5 41.Ra8 Kg7 42.Rdd8 Bc6 43.Rg8+ Kh6 
  White has a decisive advantage: +- (2.91 ++) (98% +95% =5% -0%)  Depth: 48/62   00:04:56  5643MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Re7 25.Red1 Rc7 26.Ng5 Qc8 27.b5 b6 28.Rd6 Rb7 29.Kh2 Rc7 30.f3 Rb7 31.Kg3 Rc7 32.Nh3 Qb8 33.Nxd7 Rcxd7 34.Rxd7 Rxd7 35.Rxd7 Qxe5+ 36.f4 Qe1+ 37.Kh2 Qh4 38.Bg5 Qh5 39.Bf6+ Kg8 40.Rxa7 h6 41.Rg7+ Kf8 42.Rh7 Kg8 43.Rh8+ Kf7 44.Be5 Ke7 45.Bc7 Qe2 46.Rxh6 Qxb5 47.Rh7+ Ke8 48.Ng5 Qb4 49.Be5 Qb3 50.Bf6 Qb4 51.Nxe6 Qd6 52.Nc7+ Kf8 53.Be7+ Qxe7 54.Rxe7 
  White has a decisive advantage: +- (3.17) (99% +97% =3% -0%)  Depth: 48/68   00:06:51  7821MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Re7 25.Red1 Rc7 26.Ng5 Qc8 27.b5 b6 28.Rd6 Rb7 29.Kh2 Rc7 30.f3 Rb7 31.Kg3 Rc7 32.Nh3 Qb8 33.Nxd7 Rcxd7 34.Rxd7 Rxd7 35.Rxd7 Qxe5+ 36.f4 Qe1+ 37.Kh2 Qh4 38.Bg5 Qh5 39.Bf6+ Kg8 40.Rxa7 h6 41.Rg7+ Kf8 42.Rh7 Kg8 43.Rh8+ Kf7 44.Be5 Ke7 45.Bc7 Qe2 46.Rxh6 Qxb5 47.Rh7+ Ke8 48.Ng5 Qb4 49.Be5 Qb3 50.Bf6 Qb4 51.Nxe6 Qd6 52.Nc7+ Kf8 53.Be7+ Qxe7 54.Rxe7 
  White has a decisive advantage: +- (3.25 ++) (99% +98% =2% -0%)  Depth: 49/72   00:09:57  11349MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Re7 25.Red1 Rc7 26.Ng5 Qc8 27.b5 b6 28.Rd6 Rb7 29.Kh2 Rc7 30.f3 Rb7 31.Kg3 Rc7 32.Nh3 Qb8 33.Nxd7 Rcxd7 34.Rxd7 Rxd7 35.Rxd7 Qxe5+ 36.f4 Qe1+ 37.Kh2 Qh4 38.Bg5 Qh5 39.Bf6+ Kg8 40.Rxa7 h6 41.Rg7+ Kf8 42.Rh7 Kg8 43.Rh8+ Kf7 44.Be5 Ke7 45.Bc7 Qe2 46.Rxh6 Qxb5 47.Rh7+ Ke8 48.Ng5 Qb4 49.Be5 Qb3 50.Bf6 Qb4 51.Nxe6 Qd6 52.Nc7+ Kf8 53.Be7+ Qxe7 54.Rxe7 
  White has a decisive advantage: +- (3.34 ++) (99% +98% =2% -0%)  Depth: 49/72   00:10:49  12367MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Re7 25.Red1 Rc7 26.Ng5 Qc8 27.b5 b6 28.Rd6 Rb7 29.Kh2 Rc7 30.f3 Rb7 31.Kg3 Rc7 32.Nh3 Qb8 33.Nxd7 Rcxd7 34.Rxd7 Rxd7 35.Rxd7 Qxe5+ 36.f4 Qe1+ 37.Kh2 Qh4 38.Bg5 Qh5 39.Bf6+ Kg8 40.Rxa7 h6 41.Rg7+ Kf8 42.Rh7 Kg8 43.Rh8+ Kf7 44.Be5 Ke7 45.Bc7 Qe2 46.Rxh6 Qxb5 47.Rh7+ Ke8 48.Ng5 Qb4 49.Be5 Qb3 50.Bf6 Qb4 51.Nxe6 Qd6 52.Nc7+ Kf8 53.Be7+ Qxe7 54.Rxe7 
  White has a decisive advantage: +- (3.46 ++) (99% +98% =2% -0%)  Depth: 49/72   00:13:14  15188MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Re7 25.Red1 Rc7 26.Ng5 Qc8 27.b5 b6 28.Rd6 Rb7 29.Kh2 Rc7 30.f3 Rb7 31.Kg3 Rc7 32.Nh3 Qb8 33.Nxd7 Rcxd7 34.Rxd7 Rxd7 35.Rxd7 Qxe5+ 36.f4 Qe1+ 37.Kh2 Qh4 38.Bg7+ Kg8 39.Be5 Qh5 40.Rxa7 h6 41.Ra8+ Kf7 42.Rh8 Ke7 43.Bc7 e5 44.Bxe5 Kf7 45.Bc3 Qh4 46.Rh7+ Ke6 47.Bd4 Qd8 48.Be3 Kd5 49.Rxh6 Qe7 50.Bxb6 Qg7 51.Rh4 Qb7 52.Bf2 Qxb5 53.Ng5 Qb4 
  White has a decisive advantage: +- (3.41) (99% +98% =2% -0%)  Depth: 49/73   00:15:40  17994MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Re7 25.Red1 Rc7 26.Ng5 Qc8 27.b5 b6 28.Rd6 Rb7 29.Kh2 Rc7 30.f3 Rb7 31.Kg3 Rc7 32.Nh3 Qb8 33.Nxd7 Rcxd7 34.Rxd7 Rxd7 35.Rxd7 Qxe5+ 36.f4 Qe1+ 37.Kh2 Qh4 38.Bg7+ Kg8 39.Be5 Qh5 40.Rxa7 h6 41.Ra8+ Kf7 42.Rh8 Ke7 43.Bc7 e5 44.Bxe5 Kf7 45.Bc3 Qh4 46.Rh7+ Ke6 47.Bd4 Qd8 48.Be3 Kd5 49.Rxh6 Qe7 50.Bxb6 Qg7 51.Rh4 Qb7 52.Bf2 Qxb5 53.Ng5 Qb4 
  White has a decisive advantage: +- (3.50 ++) (99% +99% =2% -0%)  Depth: 50/75   00:17:00  19478MN

Analysis by Fat Titz 260821 64 BMI2 NUMA:

Code: Select all

...
21.Qxd4 Nd5 22.Nd6 Bxd6 23.exd6 f6 24.Ng5 Nb6 25.Ne4 Kf7 26.Ng5+ 
  The position is equal: = (0.00)  Depth: 45/12   00:01:24  782MN
21.Qxb4 
  The position is equal: = (0.08 ++)  Depth: 46/12   00:02:33  1410MN
21.Qxb4 
  The position is equal: = (0.16 ++)  Depth: 46/12   00:02:42  1497MN
21.Qxb4 
  White has an edge: = (0.29 ++)  Depth: 46/12   00:02:54  1619MN
21.Qxb4 
  White is slightly better: +/= (0.47 ++)  Depth: 46/12   00:03:04  1712MN
21.Qxb4 
  White is better: +/- (0.71 ++)  Depth: 46/12   00:03:18  1851MN
21.Qxb4 
  White is better: +/- (1.04 ++)  Depth: 46/19   00:03:43  2105MN
21.Qxb4 
  White is clearly better: +/- (1.47 ++)  Depth: 46/19   00:04:10  2368MN
21.Qxb4 
  White is clearly better: +- (2.03 ++)  Depth: 46/31   00:04:45  2722MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Rc7 25.Red1 Re7 26.Ng5 Qc8 27.b5 b6 28.Kh2 Rb7 29.R1d3 Rc7 30.Nh3 Rb7 31.Nf4 Qc7 32.Rxd7 Rxd7 33.Rxd7 Qxd7 34.Nxd7 Rxd7 35.Nxe6 Rd5 36.Bf4 Rd7 37.Ng5 h6 38.e6 Re7 39.Nf7+ Kg8 40.Nd8 Re8 41.Bc7 Kg7 42.Kg3 Kf6 43.Kf4 Rf8 44.Kf3 Ke7 45.Nc6+ Kxe6 46.Nxa7 Ra8 47.Bxb6 
  White is clearly better: +- (2.35)  Depth: 46/65   00:05:17  3026MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Rc7 25.Red1 Re7 26.Ng5 Qc8 27.b5 b6 28.Kh2 Rb7 29.R1d3 Rc7 30.Nh3 Rb7 31.Nf4 Qc7 32.Rxd7 Rxd7 33.Rxd7 Qxd7 34.Nxd7 Rxd7 35.Nxe6 Rd5 36.Bf4 Rd7 37.Ng5 h6 38.e6 Re7 39.Nf7+ Kg8 40.Nd8 Re8 41.Bc7 Kg7 42.Kg3 Kf6 43.Kf4 Rf8 44.Kf3 Ke7 45.Nc6+ Kxe6 46.Nxa7 Ra8 47.Bxb6 
  White is clearly better: +- (2.43 ++)  Depth: 47/45   00:08:33  4878MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Rc7 25.Red1 Re7 26.Ng5 Qc8 27.b5 b6 28.Kh2 Rb7 29.R1d3 Rc7 30.Nh3 Rb7 31.Nf4 Qc7 32.Rxd7 Rxd7 33.Rxd7 Qxd7 34.Nxd7 Rxd7 35.Nxe6 Rd5 36.Bf4 Rd7 37.Ng5 h6 38.e6 Re7 39.Nf7+ Kg8 40.Nd8 Re8 41.Bc7 Kg7 42.Kg3 Kf6 43.Kf4 Rf8 44.Kf3 Ke7 45.Nc6+ Kxe6 46.Nxa7 Ra8 47.Bxb6 
  White has a decisive advantage: +- (2.51 ++)  Depth: 47/45   00:09:15  5266MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Rc7 25.Red1 Re7 26.Ng5 Qc8 27.b5 b6 28.Kh2 Rb7 29.R1d3 Rc7 30.Nh3 Rb7 31.Nf4 Qc7 32.Rxd7 Rxd7 33.Rxd7 Qxd7 34.Nxd7 Rxd7 35.Nxe6 Rd5 36.Bf4 Rd7 37.Ng5 h6 38.e6 Re7 39.Nf7+ Kg8 40.Nd8 Re8 41.Bc7 Kg7 42.Kg3 Kf6 43.Kf4 Rf8 44.Kf3 Ke7 45.Nc6+ Kxe6 46.Nxa7 Ra8 47.Bxb6 
  White has a decisive advantage: +- (2.64 ++)  Depth: 47/45   00:09:36  5468MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Rc7 25.Red1 Re7 26.Ng5 Qc8 27.b5 b6 28.Kh2 Rb7 29.R1d3 Rc7 30.Nh3 Rb7 31.Nf4 Qc7 32.Rxd7 Rxd7 33.Rxd7 Qxd7 34.Nxd7 Rxd7 35.Nxe6 Rd5 36.Bf4 Rd7 37.Ng5 h6 38.e6 Re7 39.Nf7+ Kg8 40.Nd8 Re8 41.Bc7 Kg7 42.Kg3 Kf6 43.Kf4 Rf8 44.Kf3 Ke7 45.Nc6+ Kxe6 46.Nxa7 Ra8 47.Bxb6 
  White has a decisive advantage: +- (2.82 ++)  Depth: 47/45   00:09:47  5564MN
21.Qxb4 Bxb4 22.axb4 f5 23.Nf6+ Kh8 24.Rxd4 Rc7 25.Red1 Re7 26.Ng5 Qc8 27.b5 b6 28.Kh2 Rb7 29.R1d3 Rc7 30.Nh3 Rb7 31.Nf4 Qc7 32.Rxd7 Rxd7 33.Rxd7 Qxd7 34.Nxd7 Rxd7 35.Nxe6 Rd5 36.Bf4 Rd7 37.Ng5 h6 38.e6 Re7 39.Nf7+ Kg8 40.Nd8 Re8 41.Bc7 Kg7 42.Kg3 Kf6 43.Kf4 Rf8 44.Kf3 Ke7 45.Nc6+ Kxe6 46.Nxa7 Ra8 47.Bxb6 
  White has a decisive advantage: +- (3.06 ++)  Depth: 47/45   00:11:06  6304MN

Test position: Lc0 vs Stockfish, TCEC 18, game 65

Re: Test position: Lc0 vs Stockfish, TCEC 18, game 65

Re: Test position: Lc0 vs Stockfish, TCEC 18, game 65

Re: Test position: Lc0 vs Stockfish, TCEC 18, game 65

Re: Test position: Lc0 vs Stockfish, TCEC 18, game 65

Re: Test position: Lc0 vs Stockfish, TCEC 18, game 65

Re: Test position: Lc0 vs Stockfish, TCEC 18, game 65

Re: Test position: Lc0 vs Stockfish, TCEC 18, game 65

Re: Test position: Lc0 vs Stockfish, TCEC 18, game 65

Re: Test position: Lc0 vs Stockfish, TCEC 18, game 65

Re: Test position: Lc0 vs Stockfish, TCEC 18, game 65