Komodo vs. GM Lenderman match; "Almost a knight"

[Angrim]

It seems to me that an "obvious" way to handle these odds games would be to have the computer scale the value of it's pieces up until the initial position scores as 0.0. That way the further behind it starts, the less willing it will be to trade pieces.

[lkaufman]

It seems to me that an "obvious" way to handle these odds games would be to have the computer scale the value of it's pieces up until the initial position scores as 0.0. That way the further behind it starts, the less willing it will be to trade pieces.

Well, that's effectively what setting a high Contempt value does. However, aside from the detail that Komodo has a limit of 250 for Contempt, less than a knight, at these high handicaps if you do set Contempt high enough to show a near-zero score it just plays poorly; it is so afraid of trading that it won't initiate any action and just tries to "hide", which of course doesn't work. I've found that setting Contempt at a value that cuts the eval deficit in half works about as well as anything; it will try to avoid major trades, but won't go to extremes to avoid them. I think Komodo (probably MCTS version) plays these handicaps better with such a setting than any other engine with the possible exception of Lc0 11248 or similar networks, but still not all that well. I don't have a better solution right now though.

[Chessqueen]

It seems to me that an "obvious" way to handle these odds games would be to have the computer scale the value of it's pieces up until the initial position scores as 0.0. That way the further behind it starts, the less willing it will be to trade pieces.

Well, that's effectively what setting a high Contempt value does. However, aside from the detail that Komodo has a limit of 250 for Contempt, less than a knight, at these high handicaps if you do set Contempt high enough to show a near-zero score it just plays poorly; it is so afraid of trading that it won't initiate any action and just tries to "hide", which of course doesn't work. I've found that setting Contempt at a value that cuts the eval deficit in half works about as well as anything; it will try to avoid major trades, but won't go to extremes to avoid them. I think Komodo (probably MCTS version) plays these handicaps better with such a setting than any other engine with the possible exception of Lc0 11248 or similar networks, but still not all that well. I don't have a better solution right now though.

This was a tough game and I am not ready to take Komodo 13.3 with a Knight Odds yet, maybe in 5 more months of training with my online trainer. Larry you should play some of the Knight Odds in Chess960 and keep it private to see how you can make Komodo better to face GM Lenderman, but without him removing a pawn it would it seems impossible from my point of view for your Beast to take such as high rated GM

[pgn]
[Event "?"]
[Site "?"]
[Date "2020.05.04"]
[Round "?"]
[White "Komodo 13.3 64-bit"]
[Black "ChessQueen"]
[Result "0-1"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PP1PPPPP/R1BQKBNR w KQkq - 0 1"]
[PlyCount "68"]

1. Nf3 {[%emt 0:00:35]} d5 {[%emt 0:00:36]} 2. d4 {[%emt 0:00:34]} e6 {
[%emt 0:00:42]} 3. g3 {[%emt 0:00:42]} c5 {[%emt 0:00:06]} 4. Bg2 {
[%emt 0:00:38]} Nf6 {[%emt 0:00:09]} 5. O-O {[%emt 0:00:36]} Nc6 {
[%emt 0:00:09]} 6. e3 {[%emt 0:00:28]} cxd4 {[%emt 0:00:14]} 7. exd4 {
[%emt 0:00:24]} Bd6 {[%emt 0:00:04]} 8. a3 {[%emt 0:00:33]} Qb6 {[%emt 0:00:13]
} 9. Be3 {[%emt 0:00:35]} Qxb2 {[%emt 0:00:20]} 10. Qd3 {[%emt 0:00:22]} Qb6 {
[%emt 0:00:07]} 11. Ne5 {[%emt 0:00:26]} Bxe5 {[%emt 0:00:24]} 12. dxe5 {
[%emt 0:00:16]} Nxe5 {[%emt 0:00:06]} 13. Qc3 {[%emt 0:00:16]} Qd6 {
[%emt 0:00:28]} 14. Rfd1 {[%emt 0:00:03]} Nc6 {[%emt 0:00:22]} 15. Bc5 {
[%emt 0:00:33]} Qe5 {[%emt 0:00:22]} 16. Qc1 {[%emt 0:00:08]} Qf5 {
[%emt 0:00:24]} 17. Rb1 {[%emt 0:00:30]} Ne4 {[%emt 0:00:13]} 18. Re1 {
[%emt 0:00:28]} Qg6 {[%emt 0:00:33]} 19. a4 {[%emt 0:00:26]} b6 {[%emt 0:00:05]
} 20. Bxb6 {[%emt 0:00:17]} Bd7 {[%emt 0:00:23]} 21. Bc5 {[%emt 0:00:01]} Ne5 {
[%emt 0:00:07]} 22. Qa3 {[%emt 0:00:15]} a5 {[%emt 0:00:16]} 23. Bd4 {
[%emt 0:00:16]} Nc4 {[%emt 0:00:11]} 24. Qa2 {[%emt 0:00:09]} f5 {
[%emt 0:00:15]} 25. Rb7 {[%emt 0:00:10]} Qf7 {[%emt 0:00:19]} 26. Bf1 {
[%emt 0:00:30]} O-O {[%emt 0:00:06]} 27. Qc2 {[%emt 0:00:17]} e5 {
[%emt 0:00:22]} 28. Bxc4 {[%emt 0:00:14]} dxc4 {[%emt 0:00:07]} 29. Bxe5 {
[%emt 0:00:15]} Nc5 {[%emt 0:00:18]} 30. Rb6 {[%emt 0:00:20]} Nd3 {
[%emt 0:00:13]} 31. Re3 {[%emt 0:00:15]} Rae8 {[%emt 0:00:08]} 32. Rxd3 {
[%emt 0:00:09]} cxd3 {[%emt 0:00:17]} 33. Qc3 {[%emt 0:00:01]} d2 {
[%emt 0:00:09]} 34. Rd6 {[%emt 0:00:04]} Rxe5 {[%emt 0:00:12]} 0-1[/pgn]

[JJJ]

It seems to me that an "obvious" way to handle these odds games would be to have the computer scale the value of it's pieces up until the initial position scores as 0.0. That way the further behind it starts, the less willing it will be to trade pieces.

Well, that's effectively what setting a high Contempt value does. However, aside from the detail that Komodo has a limit of 250 for Contempt, less than a knight, at these high handicaps if you do set Contempt high enough to show a near-zero score it just plays poorly; it is so afraid of trading that it won't initiate any action and just tries to "hide", which of course doesn't work. I've found that setting Contempt at a value that cuts the eval deficit in half works about as well as anything; it will try to avoid major trades, but won't go to extremes to avoid them. I think Komodo (probably MCTS version) plays these handicaps better with such a setting than any other engine with the possible exception of Lc0 11248 or similar networks, but still not all that well. I don't have a better solution right now though.

So Larry, will you train a NN komodo to play any sort of handicap vs human and hoping to have a better net than Lc0 ?

Jean-Baptiste.

[lkaufman]

It seems to me that an "obvious" way to handle these odds games would be to have the computer scale the value of it's pieces up until the initial position scores as 0.0. That way the further behind it starts, the less willing it will be to trade pieces.

Well, that's effectively what setting a high Contempt value does. However, aside from the detail that Komodo has a limit of 250 for Contempt, less than a knight, at these high handicaps if you do set Contempt high enough to show a near-zero score it just plays poorly; it is so afraid of trading that it won't initiate any action and just tries to "hide", which of course doesn't work. I've found that setting Contempt at a value that cuts the eval deficit in half works about as well as anything; it will try to avoid major trades, but won't go to extremes to avoid them. I think Komodo (probably MCTS version) plays these handicaps better with such a setting than any other engine with the possible exception of Lc0 11248 or similar networks, but still not all that well. I don't have a better solution right now though.

So Larry, will you train a NN komodo to play any sort of handicap vs human and hoping to have a better net than Lc0 ?

Jean-Baptiste.

Well, I'd like to do that. But Mark and I don't know much about training NNs, we rely on chess.com guys for that, and we need to have a marketable Komodo NN version before we can try to optimize it in such ways.

[lkaufman]

It seems to me that an "obvious" way to handle these odds games would be to have the computer scale the value of it's pieces up until the initial position scores as 0.0. That way the further behind it starts, the less willing it will be to trade pieces.

Well, that's effectively what setting a high Contempt value does. However, aside from the detail that Komodo has a limit of 250 for Contempt, less than a knight, at these high handicaps if you do set Contempt high enough to show a near-zero score it just plays poorly; it is so afraid of trading that it won't initiate any action and just tries to "hide", which of course doesn't work. I've found that setting Contempt at a value that cuts the eval deficit in half works about as well as anything; it will try to avoid major trades, but won't go to extremes to avoid them. I think Komodo (probably MCTS version) plays these handicaps better with such a setting than any other engine with the possible exception of Lc0 11248 or similar networks, but still not all that well. I don't have a better solution right now though.

This was a tough game and I am not ready to take Komodo 13.3 with a Knight Odds yet, maybe in 5 more months of training with my online trainer. Larry you should play some of the Knight Odds in Chess960 and keep it private to see how you can make Komodo better to face GM Lenderman, but without him removing a pawn it would it seems impossible from my point of view for your Beast to take such as high rated GM

[pgn]
[Event "?"]
[Site "?"]
[Date "2020.05.04"]
[Round "?"]
[White "Komodo 13.3 64-bit"]
[Black "ChessQueen"]
[Result "0-1"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PP1PPPPP/R1BQKBNR w KQkq - 0 1"]
[PlyCount "68"]

1. Nf3 {[%emt 0:00:35]} d5 {[%emt 0:00:36]} 2. d4 {[%emt 0:00:34]} e6 {
[%emt 0:00:42]} 3. g3 {[%emt 0:00:42]} c5 {[%emt 0:00:06]} 4. Bg2 {
[%emt 0:00:38]} Nf6 {[%emt 0:00:09]} 5. O-O {[%emt 0:00:36]} Nc6 {
[%emt 0:00:09]} 6. e3 {[%emt 0:00:28]} cxd4 {[%emt 0:00:14]} 7. exd4 {
[%emt 0:00:24]} Bd6 {[%emt 0:00:04]} 8. a3 {[%emt 0:00:33]} Qb6 {[%emt 0:00:13]
} 9. Be3 {[%emt 0:00:35]} Qxb2 {[%emt 0:00:20]} 10. Qd3 {[%emt 0:00:22]} Qb6 {
[%emt 0:00:07]} 11. Ne5 {[%emt 0:00:26]} Bxe5 {[%emt 0:00:24]} 12. dxe5 {
[%emt 0:00:16]} Nxe5 {[%emt 0:00:06]} 13. Qc3 {[%emt 0:00:16]} Qd6 {
[%emt 0:00:28]} 14. Rfd1 {[%emt 0:00:03]} Nc6 {[%emt 0:00:22]} 15. Bc5 {
[%emt 0:00:33]} Qe5 {[%emt 0:00:22]} 16. Qc1 {[%emt 0:00:08]} Qf5 {
[%emt 0:00:24]} 17. Rb1 {[%emt 0:00:30]} Ne4 {[%emt 0:00:13]} 18. Re1 {
[%emt 0:00:28]} Qg6 {[%emt 0:00:33]} 19. a4 {[%emt 0:00:26]} b6 {[%emt 0:00:05]
} 20. Bxb6 {[%emt 0:00:17]} Bd7 {[%emt 0:00:23]} 21. Bc5 {[%emt 0:00:01]} Ne5 {
[%emt 0:00:07]} 22. Qa3 {[%emt 0:00:15]} a5 {[%emt 0:00:16]} 23. Bd4 {
[%emt 0:00:16]} Nc4 {[%emt 0:00:11]} 24. Qa2 {[%emt 0:00:09]} f5 {
[%emt 0:00:15]} 25. Rb7 {[%emt 0:00:10]} Qf7 {[%emt 0:00:19]} 26. Bf1 {
[%emt 0:00:30]} O-O {[%emt 0:00:06]} 27. Qc2 {[%emt 0:00:17]} e5 {
[%emt 0:00:22]} 28. Bxc4 {[%emt 0:00:14]} dxc4 {[%emt 0:00:07]} 29. Bxe5 {
[%emt 0:00:15]} Nc5 {[%emt 0:00:18]} 30. Rb6 {[%emt 0:00:20]} Nd3 {
[%emt 0:00:13]} 31. Re3 {[%emt 0:00:15]} Rae8 {[%emt 0:00:08]} 32. Rxd3 {
[%emt 0:00:09]} cxd3 {[%emt 0:00:17]} 33. Qc3 {[%emt 0:00:01]} d2 {
[%emt 0:00:09]} 34. Rd6 {[%emt 0:00:04]} Rxe5 {[%emt 0:00:12]} 0-1[/pgn]

I think that you are right, and I agree that the four FRC games with Lenderman at full knight odds will be difficult for Komodo (I'd be happy with 2 draws out of those four games), but I think you underestimate how huge the difference is between your hardware and the threadripper, and also Komodo is improved a bit already. I do want to try to find settings or modifications that will help vs. Lenderman. One theory I have is that a lower Dynamism setting is appropriate vs. humans, since they aren't so good at exploiting mobility as machines are. So you could try playing with Dynamism reduced to perhaps 80 or so to see if that helps.

[Chessqueen]

I think that you are right, and I agree that the four FRC games with Lenderman at full knight odds will be difficult for Komodo (I'd be happy with 2 draws out of those four games), but I think you underestimate how huge the difference is between your hardware and the threadripper, and also Komodo is improved a bit already. I do want to try to find settings or modifications that will help vs. Lenderman. One theory I have is that a lower Dynamism setting is appropriate vs. humans, since they aren't so good at exploiting mobility as machines are. So you could try playing with Dynamism reduced to perhaps 80 or so to see if that helps.

Dynamism =80 Since I was a piece up I saw that it would be great to return it and to open up the King 16....Bxg3+

[pgn]
[Event "?"]
[Site "?"]
[Date "2020.05.05"]
[Round "?"]
[White "Komodo 13.3 64-bit"]
[Black "ChessQueen"]
[Result "0-1"]
[SetUp "1"]
[FEN "rnbqkbnr/ppp1pppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - 0 1"]
[PlyCount "55"]

1. e4 {[%emt 0:00:31]} Nf6 {[%emt 0:01:24]} 2. e5 {[%emt 0:00:16]} Nd5 {
[%emt 0:00:08]} 3. d4 {[%emt 0:00:27]} Bf5 {[%emt 0:00:08]} 4. Bb5+ {
[%emt 0:00:22]} c6 {[%emt 0:00:07]} 5. Bd3 {[%emt 0:00:20]} Bxd3 {
[%emt 0:00:14]} 6. Qxd3 {[%emt 0:00:22]} e6 {[%emt 0:01:01]} 7. Nf3 {
[%emt 0:00:03]} c5 {[%emt 0:00:19]} 8. O-O {[%emt 0:00:30]} Nc6 {[%emt 0:00:10]
} 9. dxc5 {[%emt 0:00:09]} Bxc5 {[%emt 0:00:25]} 10. Qc4 {[%emt 0:00:02]} Qb6 {
[%emt 0:00:17]} 11. Qg4 {[%emt 0:00:13]} Nd4 {[%emt 0:00:07]} 12. Nxd4 {
[%emt 0:00:36]} Bxd4 {[%emt 0:00:08]} 13. Qxg7 {[%emt 0:00:18]} O-O-O {
[%emt 0:00:09]} 14. c3 {[%emt 0:00:27]} Rhg8 {[%emt 0:00:45]} 15. Qh6 {
[%emt 0:00:00]} Bxe5 {[%emt 0:00:28]} 16. g3 {[%emt 0:00:39]} Bxg3 {
[%emt 0:00:36]} 17. hxg3 {[%emt 0:00:18]} Rxg3+ {[%emt 0:00:05]} 18. Kh2 {
[%emt 0:00:32]} Rg4 {[%emt 0:01:38]} 19. Bd2 {[%emt 0:00:32]} Rdg8 {
[%emt 0:00:10]} 20. Qh3 {[%emt 0:00:14]} Qd6+ {[%emt 0:00:20]} 0-1[/pgn]

[lkaufman]

I think that you are right, and I agree that the four FRC games with Lenderman at full knight odds will be difficult for Komodo (I'd be happy with 2 draws out of those four games), but I think you underestimate how huge the difference is between your hardware and the threadripper, and also Komodo is improved a bit already. I do want to try to find settings or modifications that will help vs. Lenderman. One theory I have is that a lower Dynamism setting is appropriate vs. humans, since they aren't so good at exploiting mobility as machines are. So you could try playing with Dynamism reduced to perhaps 80 or so to see if that helps.

Dynamism =80 Since I was a piece up I saw that it would be great to return it and to open up the King 16....Bxg3+

[pgn]
[Event "?"]
[Site "?"]
[Date "2020.05.05"]
[Round "?"]
[White "Komodo 13.3 64-bit"]
[Black "ChessQueen"]
[Result "0-1"]
[SetUp "1"]
[FEN "rnbqkbnr/ppp1pppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - 0 1"]
[PlyCount "55"]

1. e4 {[%emt 0:00:31]} Nf6 {[%emt 0:01:24]} 2. e5 {[%emt 0:00:16]} Nd5 {
[%emt 0:00:08]} 3. d4 {[%emt 0:00:27]} Bf5 {[%emt 0:00:08]} 4. Bb5+ {
[%emt 0:00:22]} c6 {[%emt 0:00:07]} 5. Bd3 {[%emt 0:00:20]} Bxd3 {
[%emt 0:00:14]} 6. Qxd3 {[%emt 0:00:22]} e6 {[%emt 0:01:01]} 7. Nf3 {
[%emt 0:00:03]} c5 {[%emt 0:00:19]} 8. O-O {[%emt 0:00:30]} Nc6 {[%emt 0:00:10]
} 9. dxc5 {[%emt 0:00:09]} Bxc5 {[%emt 0:00:25]} 10. Qc4 {[%emt 0:00:02]} Qb6 {
[%emt 0:00:17]} 11. Qg4 {[%emt 0:00:13]} Nd4 {[%emt 0:00:07]} 12. Nxd4 {
[%emt 0:00:36]} Bxd4 {[%emt 0:00:08]} 13. Qxg7 {[%emt 0:00:18]} O-O-O {
[%emt 0:00:09]} 14. c3 {[%emt 0:00:27]} Rhg8 {[%emt 0:00:45]} 15. Qh6 {
[%emt 0:00:00]} Bxe5 {[%emt 0:00:28]} 16. g3 {[%emt 0:00:39]} Bxg3 {
[%emt 0:00:36]} 17. hxg3 {[%emt 0:00:18]} Rxg3+ {[%emt 0:00:05]} 18. Kh2 {
[%emt 0:00:32]} Rg4 {[%emt 0:01:38]} 19. Bd2 {[%emt 0:00:32]} Rdg8 {
[%emt 0:00:10]} 20. Qh3 {[%emt 0:00:14]} Qd6+ {[%emt 0:00:20]} 0-1[/pgn]

Okay, this is a bit worrysome, even on your computer I didn't expect you to win any games at knight for pawn, regardless of settings. But anyway this does convince me not to try Dynamism as low as 80, the normal value will see such sacrifices much more quickly. It would help if you always mention whether games are standard or MCTS mode, and what Contempt was used. If you find that certain settings make it harder for you to score, let us know.

[Chessqueen]

I think that you are right, and I agree that the four FRC games with Lenderman at full knight odds will be difficult for Komodo (I'd be happy with 2 draws out of those four games), but I think you underestimate how huge the difference is between your hardware and the threadripper, and also Komodo is improved a bit already. I do want to try to find settings or modifications that will help vs. Lenderman. One theory I have is that a lower Dynamism setting is appropriate vs. humans, since they aren't so good at exploiting mobility as machines are. So you could try playing with Dynamism reduced to perhaps 80 or so to see if that helps.

Dynamism =80 Since I was a piece up I saw that it would be great to return it and to open up the King 16....Bxg3+

[pgn]
[Event "?"]
[Site "?"]
[Date "2020.05.05"]
[Round "?"]
[White "Komodo 13.3 64-bit"]
[Black "ChessQueen"]
[Result "0-1"]
[SetUp "1"]
[FEN "rnbqkbnr/ppp1pppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - 0 1"]
[PlyCount "55"]

1. e4 {[%emt 0:00:31]} Nf6 {[%emt 0:01:24]} 2. e5 {[%emt 0:00:16]} Nd5 {
[%emt 0:00:08]} 3. d4 {[%emt 0:00:27]} Bf5 {[%emt 0:00:08]} 4. Bb5+ {
[%emt 0:00:22]} c6 {[%emt 0:00:07]} 5. Bd3 {[%emt 0:00:20]} Bxd3 {
[%emt 0:00:14]} 6. Qxd3 {[%emt 0:00:22]} e6 {[%emt 0:01:01]} 7. Nf3 {
[%emt 0:00:03]} c5 {[%emt 0:00:19]} 8. O-O {[%emt 0:00:30]} Nc6 {[%emt 0:00:10]
} 9. dxc5 {[%emt 0:00:09]} Bxc5 {[%emt 0:00:25]} 10. Qc4 {[%emt 0:00:02]} Qb6 {
[%emt 0:00:17]} 11. Qg4 {[%emt 0:00:13]} Nd4 {[%emt 0:00:07]} 12. Nxd4 {
[%emt 0:00:36]} Bxd4 {[%emt 0:00:08]} 13. Qxg7 {[%emt 0:00:18]} O-O-O {
[%emt 0:00:09]} 14. c3 {[%emt 0:00:27]} Rhg8 {[%emt 0:00:45]} 15. Qh6 {
[%emt 0:00:00]} Bxe5 {[%emt 0:00:28]} 16. g3 {[%emt 0:00:39]} Bxg3 {
[%emt 0:00:36]} 17. hxg3 {[%emt 0:00:18]} Rxg3+ {[%emt 0:00:05]} 18. Kh2 {
[%emt 0:00:32]} Rg4 {[%emt 0:01:38]} 19. Bd2 {[%emt 0:00:32]} Rdg8 {
[%emt 0:00:10]} 20. Qh3 {[%emt 0:00:14]} Qd6+ {[%emt 0:00:20]} 0-1[/pgn]

Okay, this is a bit worrysome, even on your computer I didn't expect you to win any games at knight for pawn, regardless of settings. But anyway this does convince me not to try Dynamism as low as 80, the normal value will see such sacrifices much more quickly. It would help if you always mention whether games are standard or MCTS mode, and what Contempt was used. If you find that certain settings make it harder for you to score, let us know.

I was very Lucky to win this one , since I only had 5 seconds left on the clock

[pgn]
[Event "Blitz:15'+10""]
[Site "MyTown"]
[Date "????.??.??"]
[Round "?"]
[White "Komodo 13.3 64-bit"]
[Black "ChessQueen"]
[Result "0-1"]
[SetUp "1"]
[FEN "1rbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/1NBQKBNR w Kk - 0 1"]
[PlyCount "82"]


1. c4 {[%emt 0:04:51]} e6 {[%emt 0:00:20]} 2. e3 {[%emt 0:02:58]} b6 {
[%emt 0:00:00]} 3. Be2 {[%emt 0:02:23]} Bb7 {[%emt 0:00:00]} 4. Nf3 {
[%emt 0:01:20]} Nf6 {[%emt 0:00:06]} 5. O-O {[%emt 0:00:18]} Be7 {
[%emt 0:00:00]} 6. h3 {[%emt 0:00:41]} c5 {[%emt 0:00:35]} 7. d3 {
[%emt 0:00:58]} Qc7 {[%emt 0:00:23]} 8. b3 {[%emt 0:00:44]} d5 {[%emt 0:00:06]}
9. Bb2 {[%emt 0:00:48]} O-O {[%emt 0:00:06]} 10. Nbd2 {[%emt 0:00:49]} dxc4 {
[%emt 0:00:06]} 11. bxc4 {[%emt 0:00:51]} Nd7 {[%emt 0:00:08]} 12. Qa1 {
[%emt 0:00:54]} Bf6 {[%emt 0:00:15]} 13. d4 {[%emt 0:00:41]} Rfe8 {
[%emt 0:00:18]} 14. Rd1 {[%emt 0:00:37]} Rbd8 {[%emt 0:00:42]} 15. Bd3 {
[%emt 0:00:33]} cxd4 {[%emt 0:00:11]} 16. exd4 {[%emt 0:00:35]} e5 {
[%emt 0:00:05]} 17. d5 {[%emt 0:00:30]} Nc5 {[%emt 0:01:06]} 18. Bf1 {
[%emt 0:01:06]} b5 {[%emt 0:00:12]} 19. Rc1 {[%emt 0:00:35]} Qe7 {
[%emt 0:00:11]} 20. Rc2 {[%emt 0:00:24]} bxc4 {[%emt 0:00:32]} 21. d6 {
[%emt 0:00:03]} Rxd6 {[%emt 0:00:23]} 22. Nxc4 {[%emt 0:00:15]} Re6 {
[%emt 0:00:11]} 23. Nh2 {[%emt 0:00:22]} e4 {[%emt 0:00:07]} 24. Bxf6 {
[%emt 0:00:38]} Rxf6 {[%emt 0:00:07]} 25. Ng4 {[%emt 0:00:33]} Rg6 {
[%emt 0:00:06]} 26. Qd4 {[%emt 0:00:31]} Rd8 {[%emt 0:00:18]} 27. Qe5 {
[%emt 0:00:12]} Re6 {[%emt 0:00:06]} 28. Qc3 {[%emt 0:00:21]} h5 {
[%emt 0:00:08]} 29. Nge3 {[%emt 0:00:29]} Nd3 {[%emt 0:00:07]} 30. Qa5 {
[%emt 0:00:25]} g6 {[%emt 0:00:14]} 31. Rd2 {[%emt 0:00:12]} f5 {[%emt 0:00:17]
} 32. g3 {[%emt 0:00:29]} h4 {[%emt 0:00:15]} 33. gxh4 {[%emt 0:00:16]} Rc8 {
[%emt 0:00:28]} 34. Qa4 {[%emt 0:00:09]} f4 {[%emt 0:00:22]} 35. Ng4 {
[%emt 0:00:27]} e3 {[%emt 0:00:11]} 36. fxe3 {[%emt 0:00:25]} Qxh4 {
[%emt 0:00:15]} 37. a3 {[%emt 0:00:09]} Qg3+ {[%emt 0:00:41]} 38. Bg2 {
[%emt 0:00:27]} Rb6 {[%emt 0:00:26]} 39. Qc2 {[%emt 0:00:29]} Ne1 {
[%emt 0:00:10]} 40. Qd1 {[%emt 0:00:09]} Nf3+ {[%emt 0:00:07]} 41. Kf1 {
[%emt 0:00:10]} Nxd2+ {[%emt 0:00:15]} 0-1[/pgn]

[lkaufman]

I think that you are right, and I agree that the four FRC games with Lenderman at full knight odds will be difficult for Komodo (I'd be happy with 2 draws out of those four games), but I think you underestimate how huge the difference is between your hardware and the threadripper, and also Komodo is improved a bit already. I do want to try to find settings or modifications that will help vs. Lenderman. One theory I have is that a lower Dynamism setting is appropriate vs. humans, since they aren't so good at exploiting mobility as machines are. So you could try playing with Dynamism reduced to perhaps 80 or so to see if that helps.

Dynamism =80 Since I was a piece up I saw that it would be great to return it and to open up the King 16....Bxg3+

[pgn]
[Event "?"]
[Site "?"]
[Date "2020.05.05"]
[Round "?"]
[White "Komodo 13.3 64-bit"]
[Black "ChessQueen"]
[Result "0-1"]
[SetUp "1"]
[FEN "rnbqkbnr/ppp1pppp/8/8/8/8/PPPPPPPP/R1BQKBNR w KQkq - 0 1"]
[PlyCount "55"]

1. e4 {[%emt 0:00:31]} Nf6 {[%emt 0:01:24]} 2. e5 {[%emt 0:00:16]} Nd5 {
[%emt 0:00:08]} 3. d4 {[%emt 0:00:27]} Bf5 {[%emt 0:00:08]} 4. Bb5+ {
[%emt 0:00:22]} c6 {[%emt 0:00:07]} 5. Bd3 {[%emt 0:00:20]} Bxd3 {
[%emt 0:00:14]} 6. Qxd3 {[%emt 0:00:22]} e6 {[%emt 0:01:01]} 7. Nf3 {
[%emt 0:00:03]} c5 {[%emt 0:00:19]} 8. O-O {[%emt 0:00:30]} Nc6 {[%emt 0:00:10]
} 9. dxc5 {[%emt 0:00:09]} Bxc5 {[%emt 0:00:25]} 10. Qc4 {[%emt 0:00:02]} Qb6 {
[%emt 0:00:17]} 11. Qg4 {[%emt 0:00:13]} Nd4 {[%emt 0:00:07]} 12. Nxd4 {
[%emt 0:00:36]} Bxd4 {[%emt 0:00:08]} 13. Qxg7 {[%emt 0:00:18]} O-O-O {
[%emt 0:00:09]} 14. c3 {[%emt 0:00:27]} Rhg8 {[%emt 0:00:45]} 15. Qh6 {
[%emt 0:00:00]} Bxe5 {[%emt 0:00:28]} 16. g3 {[%emt 0:00:39]} Bxg3 {
[%emt 0:00:36]} 17. hxg3 {[%emt 0:00:18]} Rxg3+ {[%emt 0:00:05]} 18. Kh2 {
[%emt 0:00:32]} Rg4 {[%emt 0:01:38]} 19. Bd2 {[%emt 0:00:32]} Rdg8 {
[%emt 0:00:10]} 20. Qh3 {[%emt 0:00:14]} Qd6+ {[%emt 0:00:20]} 0-1[/pgn]

Okay, this is a bit worrysome, even on your computer I didn't expect you to win any games at knight for pawn, regardless of settings. But anyway this does convince me not to try Dynamism as low as 80, the normal value will see such sacrifices much more quickly. It would help if you always mention whether games are standard or MCTS mode, and what Contempt was used. If you find that certain settings make it harder for you to score, let us know.

I was very Lucky to win this one , since I only had 5 seconds left on the clock

[pgn]
[Event "Blitz:15'+10""]
[Site "MyTown"]
[Date "????.??.??"]
[Round "?"]
[White "Komodo 13.3 64-bit"]
[Black "ChessQueen"]
[Result "0-1"]
[SetUp "1"]
[FEN "1rbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/1NBQKBNR w Kk - 0 1"]
[PlyCount "82"]


1. c4 {[%emt 0:04:51]} e6 {[%emt 0:00:20]} 2. e3 {[%emt 0:02:58]} b6 {
[%emt 0:00:00]} 3. Be2 {[%emt 0:02:23]} Bb7 {[%emt 0:00:00]} 4. Nf3 {
[%emt 0:01:20]} Nf6 {[%emt 0:00:06]} 5. O-O {[%emt 0:00:18]} Be7 {
[%emt 0:00:00]} 6. h3 {[%emt 0:00:41]} c5 {[%emt 0:00:35]} 7. d3 {
[%emt 0:00:58]} Qc7 {[%emt 0:00:23]} 8. b3 {[%emt 0:00:44]} d5 {[%emt 0:00:06]}
9. Bb2 {[%emt 0:00:48]} O-O {[%emt 0:00:06]} 10. Nbd2 {[%emt 0:00:49]} dxc4 {
[%emt 0:00:06]} 11. bxc4 {[%emt 0:00:51]} Nd7 {[%emt 0:00:08]} 12. Qa1 {
[%emt 0:00:54]} Bf6 {[%emt 0:00:15]} 13. d4 {[%emt 0:00:41]} Rfe8 {
[%emt 0:00:18]} 14. Rd1 {[%emt 0:00:37]} Rbd8 {[%emt 0:00:42]} 15. Bd3 {
[%emt 0:00:33]} cxd4 {[%emt 0:00:11]} 16. exd4 {[%emt 0:00:35]} e5 {
[%emt 0:00:05]} 17. d5 {[%emt 0:00:30]} Nc5 {[%emt 0:01:06]} 18. Bf1 {
[%emt 0:01:06]} b5 {[%emt 0:00:12]} 19. Rc1 {[%emt 0:00:35]} Qe7 {
[%emt 0:00:11]} 20. Rc2 {[%emt 0:00:24]} bxc4 {[%emt 0:00:32]} 21. d6 {
[%emt 0:00:03]} Rxd6 {[%emt 0:00:23]} 22. Nxc4 {[%emt 0:00:15]} Re6 {
[%emt 0:00:11]} 23. Nh2 {[%emt 0:00:22]} e4 {[%emt 0:00:07]} 24. Bxf6 {
[%emt 0:00:38]} Rxf6 {[%emt 0:00:07]} 25. Ng4 {[%emt 0:00:33]} Rg6 {
[%emt 0:00:06]} 26. Qd4 {[%emt 0:00:31]} Rd8 {[%emt 0:00:18]} 27. Qe5 {
[%emt 0:00:12]} Re6 {[%emt 0:00:06]} 28. Qc3 {[%emt 0:00:21]} h5 {
[%emt 0:00:08]} 29. Nge3 {[%emt 0:00:29]} Nd3 {[%emt 0:00:07]} 30. Qa5 {
[%emt 0:00:25]} g6 {[%emt 0:00:14]} 31. Rd2 {[%emt 0:00:12]} f5 {[%emt 0:00:17]
} 32. g3 {[%emt 0:00:29]} h4 {[%emt 0:00:15]} 33. gxh4 {[%emt 0:00:16]} Rc8 {
[%emt 0:00:28]} 34. Qa4 {[%emt 0:00:09]} f4 {[%emt 0:00:22]} 35. Ng4 {
[%emt 0:00:27]} e3 {[%emt 0:00:11]} 36. fxe3 {[%emt 0:00:25]} Qxh4 {
[%emt 0:00:15]} 37. a3 {[%emt 0:00:09]} Qg3+ {[%emt 0:00:41]} 38. Bg2 {
[%emt 0:00:27]} Rb6 {[%emt 0:00:26]} 39. Qc2 {[%emt 0:00:29]} Ne1 {
[%emt 0:00:10]} 40. Qd1 {[%emt 0:00:09]} Nf3+ {[%emt 0:00:07]} 41. Kf1 {
[%emt 0:00:10]} Nxd2+ {[%emt 0:00:15]} 0-1[/pgn]

Something is really wrong here. We had about a dozen games between Komodo and various grandmasters at slow Rapid time controls at this same rook for knight handicap, and Komodo never lost a single game, drawing about half, not even losing to Nakamura! Of course your hardware is much slower, but I can't believe that a player of 2000 USCF/1900 FIDE level could win a game this way from Komodo unless something is terribly wrong. MCTS or standard shouldn't matter. What depths was it getting? What else could explain this? This handicap is way less than knight odds, and you couldn't win at that.