Can Komodo give a Queen to a 1500 chess Engine on CCRL?

[Chessqueen]

But against another 1600 rated player Komodo lost very quickly, and I lost $25.00. This is using my Laptop Intel(R) Core(TM) i7-4770 CPU @ 3.40GHz with 15.9 GB Memory

Yes I was using the MCTS of Komodo 13.3. I would like to continue testing versus many humans with your latest beta version of Komodo DragonX. Here waiting at Malina Airport Young 13 years old Daniel was looking at me going over some games with my Laptop and I asked him if he has a rating he is 1465.

[pgn][Event "Queen Odds"]
[Site "Manila Airport"]
[Date "2021.09.05"]
[Round "3"]
[White "Stockfish_14_x64_bmi2"]
[Black "Daniel"]
[Result "0-1"]
[BlackElo "1465"]
[Time "09:45:17"]
[WhiteElo "3480"]
[TimeControl "300+3"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "43"]
[WhiteType "program"]
[BlackType "human"]

1. e3 e5 2. Nc3 Nf6 3. d4 e4 4. f3 Bb4 5. a3 Bxc3+ 6. bxc3 d5 7. a4 O-O 8.
Ne2 exf3 9. gxf3 Be6 10. Nf4 Qd7 11. Nxe6 fxe6 12. Bd3 Qf7 13. O-O Qh5 14.
Kh1 Ne4 15. fxe4 Rxf1+ 16. Bxf1 Qf3+ 17. Kg1 Nd7 18. Ba3 Qxe3+ 19. Kh1 Qxc3
20. Rb1 Qxa3 21. Bg2 {White resigns}[/pgn]

With large handicaps the exact version of Dragon doesn't matter, nor does the hardware if you don't use MCTS which you shouldn't at queen odds. The only thing that matters is to set Contempt properly; for queen odds I would go with the maximum value of 250.

I'd say 11.Nxe6 is bad, when down by so many pieces doesn't make sense to trade pieces in my opinion, I think you would want to keep as many pieces as possible and try hard to complicate the position and hope the opponent keeps blundering.

You have just identified a basic difference between Stockfish and Komodo. Stockfish has code that makes no sense to a chess master because it must have shown a tiny elo gain, that causes it to make this obvious mistake. We don't allow chessically incorrect code in Komodo even if it adds an elo point. That's one reason Komodo is so much stronger than Stockfish in handicap play, even if stockfish still comes out ahead head to head in standard chess.

Most people just simply look at what engine has the highest rating without knowing that it is not all that matter, my online chess trainer recommend us to get any version of Komodo to analyze chess positions even the free version 12.2 since it is closer to how most top GM think.

[Chessqueen]

But against another 1600 rated player Komodo lost very quickly, and I lost $25.00. This is using my Laptop Intel(R) Core(TM) i7-4770 CPU @ 3.40GHz with 15.9 GB Memory

[pgn][Event "Queen Odds"]
[Site "Philippines Chess Club"]
[Date "2021.09.05"]
[Round "2"]
[White "Komodo-13.3-64bit-bmi2"]
[Black "Jose"]
[Result "0-1"]
[BlackElo "1600"]
[Time "08:09:17"]
[WhiteElo "3470"]
[TimeControl "300+3"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "43"]
[WhiteType "program"]
[BlackType "human"]

1. d4 Nf6 2. Bf4 d5 3. Nf3 Nc6 4. e3 Bg4 5. Nbd2 Bxf3 6. gxf3 Nh5 7. Bg3
Nxg3 8. hxg3 e5 9. O-O-O exd4 10. exd4 Nxd4 11. Re1+ Be7 12. c3 Ne6 13. Re5
O-O 14. f4 Bf6 15. Reh5 h6 16. Rxh6 gxh6 17. Kd1 Bg7 18. f3 c5 19. Kc2 d4
20. Bc4 dxc3 21. bxc3 {White resigns} Qd6 22. Ne4 {White resigns} *[/pgn]

This is strange because the blunder 16. Rxh6?? makes no sense, I can't replicate it at any time or depth with K14.1 (I don't have 13.3 on my laptop). Were you perhaps using MCTS? I think MCTS just falls apart completely when down a huge amount like this. I would only use MCTS for handicap play if the handicap is not more than a rook, if the time control is Rapid rather than blitz, and if the number of threads is at least four.

You mentioned that you played several games at Queen odds vs Irina, with Komodo Dragon2, there is a huge difference when playing with Komodo 13.3 and a Mere intel i7 at 3.40 GHz 4 cores. I use the normal Komodo NOT the MCTS and contempt = 250

[pgn][Event "Computer chess game"]
[Date "2021.09.07"]
[Round "?"]
[White "Komodo-13.3-64bit-bmi2"]
[Black "Irina"]
[Result "0-1"]
[BlackElo "2000"]
[Time "06:38:28"]
[WhiteElo "3470"]
[TimeControl "900+10"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "65"]
[WhiteType "program"]
[BlackType "program"]

1. Nf3 Nf6 2. e3 e6 3. b3 Be7 4. Bb2 O-O 5. d4 d5 6. Bd3 Nc6 7. a3 Bd6 8.
Nbd2 Qe7 9. Ne5 Bxe5 10. dxe5 Nd7 11. f4 Qc5 12. O-O-O Qxe3 13. Kb1 Qxf4
14. Rhf1 Qg4 15. Nf3 Qxg2 16. Rd2 Qg4 17. Rdf2 Nc5 18. Rg1 Qh5 19. Be2 Qf5
20. Rg5 Qe4 21. Rfg2 g6 22. R5g4 Qf5 23. Nh4 Qh5 24. Rxg6+ Qxg6 25. Nxg6
fxg6 26. Bb5 Bd7 27. a4 Rf5 28. Ba3 Ne4 29. Bd3 Nxe5 30. Bxe4 dxe4 31. Bc1
Bc6 32. Be3 Kf7 33. Kb2 {White resigns} *[/pgn]

[lkaufman]

But against another 1600 rated player Komodo lost very quickly, and I lost $25.00. This is using my Laptop Intel(R) Core(TM) i7-4770 CPU @ 3.40GHz with 15.9 GB Memory

[pgn][Event "Queen Odds"]
[Site "Philippines Chess Club"]
[Date "2021.09.05"]
[Round "2"]
[White "Komodo-13.3-64bit-bmi2"]
[Black "Jose"]
[Result "0-1"]
[BlackElo "1600"]
[Time "08:09:17"]
[WhiteElo "3470"]
[TimeControl "300+3"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "43"]
[WhiteType "program"]
[BlackType "human"]

1. d4 Nf6 2. Bf4 d5 3. Nf3 Nc6 4. e3 Bg4 5. Nbd2 Bxf3 6. gxf3 Nh5 7. Bg3
Nxg3 8. hxg3 e5 9. O-O-O exd4 10. exd4 Nxd4 11. Re1+ Be7 12. c3 Ne6 13. Re5
O-O 14. f4 Bf6 15. Reh5 h6 16. Rxh6 gxh6 17. Kd1 Bg7 18. f3 c5 19. Kc2 d4
20. Bc4 dxc3 21. bxc3 {White resigns} Qd6 22. Ne4 {White resigns} *[/pgn]

This is strange because the blunder 16. Rxh6?? makes no sense, I can't replicate it at any time or depth with K14.1 (I don't have 13.3 on my laptop). Were you perhaps using MCTS? I think MCTS just falls apart completely when down a huge amount like this. I would only use MCTS for handicap play if the handicap is not more than a rook, if the time control is Rapid rather than blitz, and if the number of threads is at least four.

You mentioned that you played several games at Queen odds vs Irina, with Komodo Dragon2, there is a huge difference when playing with Komodo 13.3 and a Mere intel i7 at 3.40 GHz 4 cores. I use the normal Komodo NOT the MCTS and contempt = 250

[pgn][Event "Computer chess game"]
[Date "2021.09.07"]
[Round "?"]
[White "Komodo-13.3-64bit-bmi2"]
[Black "Irina"]
[Result "0-1"]
[BlackElo "2000"]
[Time "06:38:28"]
[WhiteElo "3470"]
[TimeControl "900+10"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "65"]
[WhiteType "program"]
[BlackType "program"]

1. Nf3 Nf6 2. e3 e6 3. b3 Be7 4. Bb2 O-O 5. d4 d5 6. Bd3 Nc6 7. a3 Bd6 8.
Nbd2 Qe7 9. Ne5 Bxe5 10. dxe5 Nd7 11. f4 Qc5 12. O-O-O Qxe3 13. Kb1 Qxf4
14. Rhf1 Qg4 15. Nf3 Qxg2 16. Rd2 Qg4 17. Rdf2 Nc5 18. Rg1 Qh5 19. Be2 Qf5
20. Rg5 Qe4 21. Rfg2 g6 22. R5g4 Qf5 23. Nh4 Qh5 24. Rxg6+ Qxg6 25. Nxg6
fxg6 26. Bb5 Bd7 27. a4 Rf5 28. Ba3 Ne4 29. Bd3 Nxe5 30. Bxe4 dxe4 31. Bc1
Bc6 32. Be3 Kf7 33. Kb2 {White resigns} *[/pgn]

Although Dragon came out comfortably ahead of Irina at queen odds in blitz, it lost by 83 elo (+65, -112, =23) in Rapid (15' + 10") in my test. The hardware makes no measurable difference for queen odds, nor does the specific version, but there is a big difference between Komodo (without NNUE) and Komodo Dragon (with NNUE). NNUE does dramatically change the program, for the better in most situations. As for Contempt, I tested the max of 250 at queen odds but it was too high; probably 200 would be about right for queen odds, 150 for rook odds, 125 for knight odds.

[Chessqueen]

Although Dragon came out comfortably ahead of Irina at queen odds in blitz, it lost by 83 elo (+65, -112, =23) in Rapid (15' + 10") in my test. The hardware makes no measurable difference for queen odds, nor does the specific version, but there is a big difference between Komodo (without NNUE) and Komodo Dragon (with NNUE). NNUE does dramatically change the program, for the better in most situations. As for Contempt, I tested the max of 250 at queen odds but it was too high; probably 200 would be about right for queen odds, 150 for rook odds, 125 for knight odds.
[/quote]

Komodo Dragon2 Vs Irina at 5'+3" with Contempt = 200
[pgn][Event "Queen Odds"]
[Date "2021.09.07"]
[Round "2"]
[White "Dragon-2-64bit-avx2"]
[Black "Irina"]
[Result "1-0"]
[BlackElo "1435"]
[Time "13:11:31"]
[WhiteElo "3550"]
[TimeControl "300+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "91"]
[WhiteType "program"]
[BlackType "program"]

1. a3 Nc6 2. d3 d5 3. e3 e5 4. Nd2 Nf6 5. b3 Bd6 6. h3 O-O 7. Ne2 Be6 8. g3
d4 9. e4 Nd7 10. Bg2 Re8 11. O-O Bc5 12. Kh2 Qf6 13. Nf3 Bd6 14. g4 Rac8
15. Bg5 Qg6 16. Ng3 Ra8 17. Bd2 Nc5 18. Rab1 Rad8 19. Ra1 Ra8 20. Nf5 Bxf5
21. gxf5 Qh5 22. b4 Nd7 23. Ng1 Nf6 24. Bf3 Qh4 25. Be2 Rad8 26. Nf3 Qh5
27. Ng1 Qh4 28. Nf3 Ng4+ 29. Kg2 Qh5 30. hxg4 Qxg4+ 31. Kh2 Ne7 32. Rh1
Nxf5 33. Rag1 Qh5+ 34. Kg2 Qg6+ 35. Kf1 Qe6 36. exf5 Qa2 37. f6 g6 38. Ng5
h5 39. Rxh5 Qb1+ 40. Be1 gxh5 41. Bxh5 Qxc2 42. Bxf7+ Kf8 43. Rh1 Qxd3+ 44.
Kg1 Qh3 45. Rxh3 Bxb4 46. Rh8# 1-0[/pgn]

[Chessqueen]

Komodo Dragon2 Vs Irina at 5'+3" with contempt =205 Based on CCRL rating I believe that if Irina is rated 1435 and Jose rated 1600 and he beat Komodo 13.3 the rating difference between engines of CCRL and human favor human at Queen Odds when comparing CCRL to Humans when we thought that CCRL is equal to 150 rating points less than Humans
[pgn][Event "Queen Odds"]
[Date "2021.09.07"]
[Round "2"]
[White "Dragon-2-64bit-avx2"]
[Black "Irina"]
[Result "1-0"]
[BlackElo "1435"]
[Time "13:11:31"]
[WhiteElo "3550"]
[TimeControl "300+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "91"]
[WhiteType "program"]
[BlackType "program"]

1. e3 Nc6 2. a3 e5 3. d3 d5 4. b3 Nf6 5. Bb2 Bg4 6. Nd2 Nd7 7. h3 Qh4 8. g3
Qh6 9. Ne2 Bxe2 10. Bxe2 Bc5 11. Nf1 Qg6 12. g4 O-O 13. b4 Bd6 14. h4 Nb6
15. h5 Qe6 16. h6 gxh6 17. Ng3 Ne7 18. Nf5 Nxf5 19. gxf5 Qxf5 20. Rh5 Qf6
21. f4 Rae8 22. O-O-O d4 23. Rf1 Qg7 24. Rhh1 Kh8 25. Rhg1 Qf6 26. fxe5 Qe6
27. exd6 Qxe3+ 28. Kb1 c6 29. Bd1 Nd5 30. d7 Re7 31. Bg4 h5 32. Bf5 Rd8 33.
Rh1 Re5 34. Rxh5 Nf6 35. Rhh1 Nxd7 36. Rxh7+ Kg8 37. Rhh1 Nf6 38. Bc1 Qe2
39. Bh6 Rxf5 40. Rxf5 Ng4 41. Rg5+ Kh7 42. Rg7+ Kh8 43. Rxf7 Qg2 44. Rh4
Rb8 45. Bg7+ Kg8 46. Re7 Qf1+ 47. Kb2 Qf7 48. Rxf7 Kxf7 49. Rxg4 Rg8 50.
Rxd4 Rxg7 51. Rd7+ Kf6 52. Rxg7 Kxg7 53. Kb3 Kf6 54. Kc4 Ke5 55. Kc5 Kf4
56. Kd6 Ke3 57. Kc7 Kd2 58. c4 Kxd3 59. c5 Kc4 60. Kxb7 Kb5 61. Kxa7 Ka4
62. Kb6 Kxa3 63. b5 cxb5 64. Kxb5 Kb3 65. c6 Kc3 66. c7 Kd4 67. c8=Q Ke4
68. Qg4+ Kd3 69. Kb4 Ke3 70. Kc3 Kf2 71. Kd3 Kf1 {Black resigns} *[/pgn]

[lkaufman]

Komodo Dragon2 Vs Irina at 5'+3" with contempt =205 Based on CCRL rating I believe that if Irina is rated 1435 and Jose rated 1600 and he beat Komodo 13.3 the rating difference between engines of CCRL and human favor human at Queen Odds when comparing CCRL to Humans when we thought that CCRL is equal to 150 rating points less than Humans
[pgn][Event "Queen Odds"]
[Date "2021.09.07"]
[Round "2"]
[White "Dragon-2-64bit-avx2"]
[Black "Irina"]
[Result "1-0"]
[BlackElo "1435"]
[Time "13:11:31"]
[WhiteElo "3550"]
[TimeControl "300+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "91"]
[WhiteType "program"]
[BlackType "program"]

1. e3 Nc6 2. a3 e5 3. d3 d5 4. b3 Nf6 5. Bb2 Bg4 6. Nd2 Nd7 7. h3 Qh4 8. g3
Qh6 9. Ne2 Bxe2 10. Bxe2 Bc5 11. Nf1 Qg6 12. g4 O-O 13. b4 Bd6 14. h4 Nb6
15. h5 Qe6 16. h6 gxh6 17. Ng3 Ne7 18. Nf5 Nxf5 19. gxf5 Qxf5 20. Rh5 Qf6
21. f4 Rae8 22. O-O-O d4 23. Rf1 Qg7 24. Rhh1 Kh8 25. Rhg1 Qf6 26. fxe5 Qe6
27. exd6 Qxe3+ 28. Kb1 c6 29. Bd1 Nd5 30. d7 Re7 31. Bg4 h5 32. Bf5 Rd8 33.
Rh1 Re5 34. Rxh5 Nf6 35. Rhh1 Nxd7 36. Rxh7+ Kg8 37. Rhh1 Nf6 38. Bc1 Qe2
39. Bh6 Rxf5 40. Rxf5 Ng4 41. Rg5+ Kh7 42. Rg7+ Kh8 43. Rxf7 Qg2 44. Rh4
Rb8 45. Bg7+ Kg8 46. Re7 Qf1+ 47. Kb2 Qf7 48. Rxf7 Kxf7 49. Rxg4 Rg8 50.
Rxd4 Rxg7 51. Rd7+ Kf6 52. Rxg7 Kxg7 53. Kb3 Kf6 54. Kc4 Ke5 55. Kc5 Kf4
56. Kd6 Ke3 57. Kc7 Kd2 58. c4 Kxd3 59. c5 Kc4 60. Kxb7 Kb5 61. Kxa7 Ka4
62. Kb6 Kxa3 63. b5 cxb5 64. Kxb5 Kb3 65. c6 Kc3 66. c7 Kd4 67. c8=Q Ke4
68. Qg4+ Kd3 69. Kb4 Ke3 70. Kc3 Kf2 71. Kd3 Kf1 {Black resigns} *[/pgn]

Those are two nice games, especially the rook sacrifice in the first one. On the general topic of simulating humans using engines in handicap games, I'm finding that I can do a pretty good job of simulating a "par" grandmaster in Rapid like this: I think that the CEGT blitz ratings for engines are at about the level such that if an engine on that list played with 2' + 1" and no ponder against a human with 15' + 10" with the same FIDE rating as the CEGT blitz rating of the engine, it would be roughly fair. The engine Benjamin 1.0 by Ed Schroder is the one that has performed best with knight odds compared to other similarly rating engines, so I'm using that. It has a 2530 CEGT blitz rating, so I'm saying that "Benjamin" playing 2' + 1" blitz is a good simulation of a human GM rated 2530 FIDE playing 15' + 10" Rapid. Benjamin beat Dragon 2 (Contempt 100) in 2' + 1" knight odds blitz by 70 elo, so Dragon 2 performed at 2460. In its two serious knight odds matches at 15' + 10" Rapid with titled human players, its performance rating in those 15 games was 2460 FIDE! Perfect agreement! Of course this must be lucky, but it suggests that this is a pretty reasonable way to simulate such matches. There were a lot of details that were different between those tests, but I'm finding that most of these things just don't matter in handicap play.

[Chessqueen]

Komodo Dragon2 Vs Irina at 5'+3" with contempt =205 Based on CCRL rating I believe that if Irina is rated 1435 and Jose rated 1600 and he beat Komodo 13.3 the rating difference between engines of CCRL and human favor human at Queen Odds when comparing CCRL to Humans when we thought that CCRL is equal to 150 rating points less than Humans
[pgn][Event "Queen Odds"]
[Date "2021.09.07"]
[Round "2"]
[White "Dragon-2-64bit-avx2"]
[Black "Irina"]
[Result "1-0"]
[BlackElo "1435"]
[Time "13:11:31"]
[WhiteElo "3550"]
[TimeControl "300+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "91"]
[WhiteType "program"]
[BlackType "program"]

1. e3 Nc6 2. a3 e5 3. d3 d5 4. b3 Nf6 5. Bb2 Bg4 6. Nd2 Nd7 7. h3 Qh4 8. g3
Qh6 9. Ne2 Bxe2 10. Bxe2 Bc5 11. Nf1 Qg6 12. g4 O-O 13. b4 Bd6 14. h4 Nb6
15. h5 Qe6 16. h6 gxh6 17. Ng3 Ne7 18. Nf5 Nxf5 19. gxf5 Qxf5 20. Rh5 Qf6
21. f4 Rae8 22. O-O-O d4 23. Rf1 Qg7 24. Rhh1 Kh8 25. Rhg1 Qf6 26. fxe5 Qe6
27. exd6 Qxe3+ 28. Kb1 c6 29. Bd1 Nd5 30. d7 Re7 31. Bg4 h5 32. Bf5 Rd8 33.
Rh1 Re5 34. Rxh5 Nf6 35. Rhh1 Nxd7 36. Rxh7+ Kg8 37. Rhh1 Nf6 38. Bc1 Qe2
39. Bh6 Rxf5 40. Rxf5 Ng4 41. Rg5+ Kh7 42. Rg7+ Kh8 43. Rxf7 Qg2 44. Rh4
Rb8 45. Bg7+ Kg8 46. Re7 Qf1+ 47. Kb2 Qf7 48. Rxf7 Kxf7 49. Rxg4 Rg8 50.
Rxd4 Rxg7 51. Rd7+ Kf6 52. Rxg7 Kxg7 53. Kb3 Kf6 54. Kc4 Ke5 55. Kc5 Kf4
56. Kd6 Ke3 57. Kc7 Kd2 58. c4 Kxd3 59. c5 Kc4 60. Kxb7 Kb5 61. Kxa7 Ka4
62. Kb6 Kxa3 63. b5 cxb5 64. Kxb5 Kb3 65. c6 Kc3 66. c7 Kd4 67. c8=Q Ke4
68. Qg4+ Kd3 69. Kb4 Ke3 70. Kc3 Kf2 71. Kd3 Kf1 {Black resigns} *[/pgn]

Those are two nice games, especially the rook sacrifice in the first one. On the general topic of simulating humans using engines in handicap games, I'm finding that I can do a pretty good job of simulating a "par" grandmaster in Rapid like this: I think that the CEGT blitz ratings for engines are at about the level such that if an engine on that list played with 2' + 1" and no ponder against a human with 15' + 10" with the same FIDE rating as the CEGT blitz rating of the engine, it would be roughly fair. The engine Benjamin 1.0 by Ed Schroder is the one that has performed best with knight odds compared to other similarly rating engines, so I'm using that. It has a 2530 CEGT blitz rating, so I'm saying that "Benjamin" playing 2' + 1" blitz is a good simulation of a human GM rated 2530 FIDE playing 15' + 10" Rapid. Benjamin beat Dragon 2 (Contempt 100) in 2' + 1" knight odds blitz by 70 elo, so Dragon 2 performed at 2460. In its two serious knight odds matches at 15' + 10" Rapid with titled human players, its performance rating in those 15 games was 2460 FIDE! Perfect agreement! Of course this must be lucky, but it suggests that this is a pretty reasonable way to simulate such matches. There were a lot of details that were different between those tests, but I'm finding that most of these things just don't matter in handicap play.

This is interesting it took Komodo Dragon2 a total of 10 versions to give a pawn odds to Komodo51r2 and be even

[pgn][Event "10th versions later Pawn odds"]
[Site "MININT-UB2PIMJ"]
[Date "2021.09.07"]
[Round "?"]
[White "Komodo51r2-64bit"]
[Black "Dragon-2-64bit-avx2"]
[Result "1/2"]
[BlackElo "3580"]
[Time "15:59:42"]
[WhiteElo "3107"]
[TimeControl "1800+15"]
[SetUp "1"]
[FEN "rnbqkbnr/pp1ppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1"]
[Termination "adjudication"]
[PlyCount "192"]
[WhiteType "program"]
[BlackType "program"]


1. e4 e5 2. Nf3 Nc6 3. Bc4 Bc5 4. c3 Nf6 5. d4 exd4 6. cxd4 Bb4+ 7. Bd2
Bxd2+ 8. Nbxd2 d5 9. exd5 Nxd5 10. Qb3 Nce7 11. Qb5+ Qd7 12. O-O Qxb5 13.
Bxb5+ Kf8 14. Rfe1 g6 15. Ne4 h6 16. Ne5 Kg7 17. Rac1 Rd8 18. h4 Bf5 19.
Bc4 Rac8 20. Nc5 Rc7 21. g4 Bc8 22. f3 b6 23. Ncd3 a5 24. Kf2 g5 25. h5 Be6
26. a3 Rb7 27. b3 Rd6 28. a4 Rc7 29. Bb5 Rb7 30. Ba6 Ra7 31. Nc4 Rd8 32.
Bb5 Rb7 33. Red1 f5 34. Re1 Bg8 35. Nde5 fxg4 36. fxg4 Nf6 37. Red1 Be6 38.
Kf3 Ned5 39. Bc6 Re7 40. Bxd5 Nxd5 41. Kg3 Rc7 42. Rd2 Nf6 43. Re1 Bd5 44.
Rdd1 Be4 45. Re2 Bd5 46. Rc2 Rf8 47. Rcc1 Rfc8 48. Re1 Ne4+ 49. Kh2 Nf6 50.
Re3 Be6 51. Kg3 Nd5 52. Ree1 Nf4 53. Rc2 Bg8 54. Kh2 Rd8 55. Rd2 Nd5 56.
Kg3 Nf4 57. Re3 Nd5 58. Rf3 Be6 59. Kg2 Nf6 60. Ne3 Rc1 61. Nf5+ Kh7 62.
Nxh6 Bd5 63. Nhf7 Rg8 64. Rd3 Nxh5 65. Kf2 Bxf3 66. Rxf3 Nf4 67. Nxg5+ Rxg5
68. Rxf4 Rg7 69. Ke2 Rgc7 70. Rf3 Kg7 71. Rd3 Kf6 72. Nc4 R1xc4 73. bxc4
Rxc4 74. Rf3+ Ke7 75. Kd3 Rxa4 76. Ke4 b5 77. g5 b4 78. g6 Ra1 79. Rf5 a4
80. g7 Rg1 81. Ra5 a3 82. Kd5 Rxg7 83. Kc4 Rg4 84. Kxb4 Rxd4+ 85. Kxa3 Rd7
86. Kb4 Ke6 87. Ra1 Kf7 88. Rf1+ Ke8 89. Ra1 Rc7 90. Re1+ Kd7 91. Rf1 Kd6
92. Rd1+ Ke5 93. Re1+ Kf6 94. Rf1+ Ke7 95. Re1+ Kf7 96. Rf1+ Kg6 97. Rg1+
Kf5 98. Rf1+ Ke4 99. Re1+ Kd3 100. Rd1+ Ke2 {User Adjudication} *[/pgn]

[lkaufman]

Komodo Dragon2 Vs Irina at 5'+3" with contempt =205 Based on CCRL rating I believe that if Irina is rated 1435 and Jose rated 1600 and he beat Komodo 13.3 the rating difference between engines of CCRL and human favor human at Queen Odds when comparing CCRL to Humans when we thought that CCRL is equal to 150 rating points less than Humans
[pgn][Event "Queen Odds"]
[Date "2021.09.07"]
[Round "2"]
[White "Dragon-2-64bit-avx2"]
[Black "Irina"]
[Result "1-0"]
[BlackElo "1435"]
[Time "13:11:31"]
[WhiteElo "3550"]
[TimeControl "300+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "91"]
[WhiteType "program"]
[BlackType "program"]

1. e3 Nc6 2. a3 e5 3. d3 d5 4. b3 Nf6 5. Bb2 Bg4 6. Nd2 Nd7 7. h3 Qh4 8. g3
Qh6 9. Ne2 Bxe2 10. Bxe2 Bc5 11. Nf1 Qg6 12. g4 O-O 13. b4 Bd6 14. h4 Nb6
15. h5 Qe6 16. h6 gxh6 17. Ng3 Ne7 18. Nf5 Nxf5 19. gxf5 Qxf5 20. Rh5 Qf6
21. f4 Rae8 22. O-O-O d4 23. Rf1 Qg7 24. Rhh1 Kh8 25. Rhg1 Qf6 26. fxe5 Qe6
27. exd6 Qxe3+ 28. Kb1 c6 29. Bd1 Nd5 30. d7 Re7 31. Bg4 h5 32. Bf5 Rd8 33.
Rh1 Re5 34. Rxh5 Nf6 35. Rhh1 Nxd7 36. Rxh7+ Kg8 37. Rhh1 Nf6 38. Bc1 Qe2
39. Bh6 Rxf5 40. Rxf5 Ng4 41. Rg5+ Kh7 42. Rg7+ Kh8 43. Rxf7 Qg2 44. Rh4
Rb8 45. Bg7+ Kg8 46. Re7 Qf1+ 47. Kb2 Qf7 48. Rxf7 Kxf7 49. Rxg4 Rg8 50.
Rxd4 Rxg7 51. Rd7+ Kf6 52. Rxg7 Kxg7 53. Kb3 Kf6 54. Kc4 Ke5 55. Kc5 Kf4
56. Kd6 Ke3 57. Kc7 Kd2 58. c4 Kxd3 59. c5 Kc4 60. Kxb7 Kb5 61. Kxa7 Ka4
62. Kb6 Kxa3 63. b5 cxb5 64. Kxb5 Kb3 65. c6 Kc3 66. c7 Kd4 67. c8=Q Ke4
68. Qg4+ Kd3 69. Kb4 Ke3 70. Kc3 Kf2 71. Kd3 Kf1 {Black resigns} *[/pgn]

Those are two nice games, especially the rook sacrifice in the first one. On the general topic of simulating humans using engines in handicap games, I'm finding that I can do a pretty good job of simulating a "par" grandmaster in Rapid like this: I think that the CEGT blitz ratings for engines are at about the level such that if an engine on that list played with 2' + 1" and no ponder against a human with 15' + 10" with the same FIDE rating as the CEGT blitz rating of the engine, it would be roughly fair. The engine Benjamin 1.0 by Ed Schroder is the one that has performed best with knight odds compared to other similarly rating engines, so I'm using that. It has a 2530 CEGT blitz rating, so I'm saying that "Benjamin" playing 2' + 1" blitz is a good simulation of a human GM rated 2530 FIDE playing 15' + 10" Rapid. Benjamin beat Dragon 2 (Contempt 100) in 2' + 1" knight odds blitz by 70 elo, so Dragon 2 performed at 2460. In its two serious knight odds matches at 15' + 10" Rapid with titled human players, its performance rating in those 15 games was 2460 FIDE! Perfect agreement! Of course this must be lucky, but it suggests that this is a pretty reasonable way to simulate such matches. There were a lot of details that were different between those tests, but I'm finding that most of these things just don't matter in handicap play.

This is interesting it took Komodo Dragon2 a total of 10 versions to give a pawn odds to Komodo51r2 and be even

[pgn][Event "10th versions later Pawn odds"]
[Site "MININT-UB2PIMJ"]
[Date "2021.09.07"]
[Round "?"]
[White "Komodo51r2-64bit"]
[Black "Dragon-2-64bit-avx2"]
[Result "1/2"]
[BlackElo "3580"]
[Time "15:59:42"]
[WhiteElo "3107"]
[TimeControl "1800+15"]
[SetUp "1"]
[FEN "rnbqkbnr/pp1ppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1"]
[Termination "adjudication"]
[PlyCount "192"]
[WhiteType "program"]
[BlackType "program"]


1. e4 e5 2. Nf3 Nc6 3. Bc4 Bc5 4. c3 Nf6 5. d4 exd4 6. cxd4 Bb4+ 7. Bd2
Bxd2+ 8. Nbxd2 d5 9. exd5 Nxd5 10. Qb3 Nce7 11. Qb5+ Qd7 12. O-O Qxb5 13.
Bxb5+ Kf8 14. Rfe1 g6 15. Ne4 h6 16. Ne5 Kg7 17. Rac1 Rd8 18. h4 Bf5 19.
Bc4 Rac8 20. Nc5 Rc7 21. g4 Bc8 22. f3 b6 23. Ncd3 a5 24. Kf2 g5 25. h5 Be6
26. a3 Rb7 27. b3 Rd6 28. a4 Rc7 29. Bb5 Rb7 30. Ba6 Ra7 31. Nc4 Rd8 32.
Bb5 Rb7 33. Red1 f5 34. Re1 Bg8 35. Nde5 fxg4 36. fxg4 Nf6 37. Red1 Be6 38.
Kf3 Ned5 39. Bc6 Re7 40. Bxd5 Nxd5 41. Kg3 Rc7 42. Rd2 Nf6 43. Re1 Bd5 44.
Rdd1 Be4 45. Re2 Bd5 46. Rc2 Rf8 47. Rcc1 Rfc8 48. Re1 Ne4+ 49. Kh2 Nf6 50.
Re3 Be6 51. Kg3 Nd5 52. Ree1 Nf4 53. Rc2 Bg8 54. Kh2 Rd8 55. Rd2 Nd5 56.
Kg3 Nf4 57. Re3 Nd5 58. Rf3 Be6 59. Kg2 Nf6 60. Ne3 Rc1 61. Nf5+ Kh7 62.
Nxh6 Bd5 63. Nhf7 Rg8 64. Rd3 Nxh5 65. Kf2 Bxf3 66. Rxf3 Nf4 67. Nxg5+ Rxg5
68. Rxf4 Rg7 69. Ke2 Rgc7 70. Rf3 Kg7 71. Rd3 Kf6 72. Nc4 R1xc4 73. bxc4
Rxc4 74. Rf3+ Ke7 75. Kd3 Rxa4 76. Ke4 b5 77. g5 b4 78. g6 Ra1 79. Rf5 a4
80. g7 Rg1 81. Ra5 a3 82. Kd5 Rxg7 83. Kc4 Rg4 84. Kxb4 Rxd4+ 85. Kxa3 Rd7
86. Kb4 Ke6 87. Ra1 Kf7 88. Rf1+ Ke8 89. Ra1 Rc7 90. Re1+ Kd7 91. Rf1 Kd6
92. Rd1+ Ke5 93. Re1+ Kf6 94. Rf1+ Ke7 95. Re1+ Kf7 96. Rf1+ Kg6 97. Rg1+
Kf5 98. Rf1+ Ke4 99. Re1+ Kd3 100. Rd1+ Ke2 {User Adjudication} *[/pgn]

Komodo 5 1 r2 is rated 3055 on the CEGT blitz list on one thread, almost exactly 500 below Dragon 2. In general at 3000 level or so an average non-edge pawn is worth about 300 elo, and here you also gave the weaker engine White ( so it's "pawn and a move" odds). I would expect Dragon 2 to win a match played that way, perhaps by a score like 7 to 3 or so, with the old engine drawing lots of games but winning few if any. If we use CEGT blitz ratings (because they use Ordo unlike CCRL which uses BayesElo which contracts rating differences), a good guide is that f7 odds is about 400 elo, f2 + c2 about 600, f7 + c7 about 700, knight odds about 1000 elo, and rook odds about 1200 elo, though the exact numbers depend on time control, threads, and specific engine strengths and weaknesses. So to be able to give knight odds in Rapid to Carlsen and break even, we would need a CEGT blitz rating of about 3850 if all of my assumptions are correct. With SF 14 on 8 threads at 3737 this seems doable, but unfortunately I don't think it's linear, the handicap value grows as the level goes up, so even 3850 won't be enough for that. The problem is that beyond a certain point, playing "better" just doesn't help win a clearly lost game.

[Chessqueen]

Komodo Dragon2 Vs Irina at 5'+3" with contempt =205 Based on CCRL rating I believe that if Irina is rated 1435 and Jose rated 1600 and he beat Komodo 13.3 the rating difference between engines of CCRL and human favor human at Queen Odds when comparing CCRL to Humans when we thought that CCRL is equal to 150 rating points less than Humans
[pgn][Event "Queen Odds"]
[Date "2021.09.07"]
[Round "2"]
[White "Dragon-2-64bit-avx2"]
[Black "Irina"]
[Result "1-0"]
[BlackElo "1435"]
[Time "13:11:31"]
[WhiteElo "3550"]
[TimeControl "300+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "91"]
[WhiteType "program"]
[BlackType "program"]

1. e3 Nc6 2. a3 e5 3. d3 d5 4. b3 Nf6 5. Bb2 Bg4 6. Nd2 Nd7 7. h3 Qh4 8. g3
Qh6 9. Ne2 Bxe2 10. Bxe2 Bc5 11. Nf1 Qg6 12. g4 O-O 13. b4 Bd6 14. h4 Nb6
15. h5 Qe6 16. h6 gxh6 17. Ng3 Ne7 18. Nf5 Nxf5 19. gxf5 Qxf5 20. Rh5 Qf6
21. f4 Rae8 22. O-O-O d4 23. Rf1 Qg7 24. Rhh1 Kh8 25. Rhg1 Qf6 26. fxe5 Qe6
27. exd6 Qxe3+ 28. Kb1 c6 29. Bd1 Nd5 30. d7 Re7 31. Bg4 h5 32. Bf5 Rd8 33.
Rh1 Re5 34. Rxh5 Nf6 35. Rhh1 Nxd7 36. Rxh7+ Kg8 37. Rhh1 Nf6 38. Bc1 Qe2
39. Bh6 Rxf5 40. Rxf5 Ng4 41. Rg5+ Kh7 42. Rg7+ Kh8 43. Rxf7 Qg2 44. Rh4
Rb8 45. Bg7+ Kg8 46. Re7 Qf1+ 47. Kb2 Qf7 48. Rxf7 Kxf7 49. Rxg4 Rg8 50.
Rxd4 Rxg7 51. Rd7+ Kf6 52. Rxg7 Kxg7 53. Kb3 Kf6 54. Kc4 Ke5 55. Kc5 Kf4
56. Kd6 Ke3 57. Kc7 Kd2 58. c4 Kxd3 59. c5 Kc4 60. Kxb7 Kb5 61. Kxa7 Ka4
62. Kb6 Kxa3 63. b5 cxb5 64. Kxb5 Kb3 65. c6 Kc3 66. c7 Kd4 67. c8=Q Ke4
68. Qg4+ Kd3 69. Kb4 Ke3 70. Kc3 Kf2 71. Kd3 Kf1 {Black resigns} *[/pgn]

Those are two nice games, especially the rook sacrifice in the first one. On the general topic of simulating humans using engines in handicap games, I'm finding that I can do a pretty good job of simulating a "par" grandmaster in Rapid like this: I think that the CEGT blitz ratings for engines are at about the level such that if an engine on that list played with 2' + 1" and no ponder against a human with 15' + 10" with the same FIDE rating as the CEGT blitz rating of the engine, it would be roughly fair. The engine Benjamin 1.0 by Ed Schroder is the one that has performed best with knight odds compared to other similarly rating engines, so I'm using that. It has a 2530 CEGT blitz rating, so I'm saying that "Benjamin" playing 2' + 1" blitz is a good simulation of a human GM rated 2530 FIDE playing 15' + 10" Rapid. Benjamin beat Dragon 2 (Contempt 100) in 2' + 1" knight odds blitz by 70 elo, so Dragon 2 performed at 2460. In its two serious knight odds matches at 15' + 10" Rapid with titled human players, its performance rating in those 15 games was 2460 FIDE! Perfect agreement! Of course this must be lucky, but it suggests that this is a pretty reasonable way to simulate such matches. There were a lot of details that were different between those tests, but I'm finding that most of these things just don't matter in handicap play.

This is interesting it took Komodo Dragon2 a total of 10 versions to give a pawn odds to Komodo51r2 and be even

[pgn][Event "10th versions later Pawn odds"]
[Site "MININT-UB2PIMJ"]
[Date "2021.09.07"]
[Round "?"]
[White "Komodo51r2-64bit"]
[Black "Dragon-2-64bit-avx2"]
[Result "1/2"]
[BlackElo "3580"]
[Time "15:59:42"]
[WhiteElo "3107"]
[TimeControl "1800+15"]
[SetUp "1"]
[FEN "rnbqkbnr/pp1ppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1"]
[Termination "adjudication"]
[PlyCount "192"]
[WhiteType "program"]
[BlackType "program"]


1. e4 e5 2. Nf3 Nc6 3. Bc4 Bc5 4. c3 Nf6 5. d4 exd4 6. cxd4 Bb4+ 7. Bd2
Bxd2+ 8. Nbxd2 d5 9. exd5 Nxd5 10. Qb3 Nce7 11. Qb5+ Qd7 12. O-O Qxb5 13.
Bxb5+ Kf8 14. Rfe1 g6 15. Ne4 h6 16. Ne5 Kg7 17. Rac1 Rd8 18. h4 Bf5 19.
Bc4 Rac8 20. Nc5 Rc7 21. g4 Bc8 22. f3 b6 23. Ncd3 a5 24. Kf2 g5 25. h5 Be6
26. a3 Rb7 27. b3 Rd6 28. a4 Rc7 29. Bb5 Rb7 30. Ba6 Ra7 31. Nc4 Rd8 32.
Bb5 Rb7 33. Red1 f5 34. Re1 Bg8 35. Nde5 fxg4 36. fxg4 Nf6 37. Red1 Be6 38.
Kf3 Ned5 39. Bc6 Re7 40. Bxd5 Nxd5 41. Kg3 Rc7 42. Rd2 Nf6 43. Re1 Bd5 44.
Rdd1 Be4 45. Re2 Bd5 46. Rc2 Rf8 47. Rcc1 Rfc8 48. Re1 Ne4+ 49. Kh2 Nf6 50.
Re3 Be6 51. Kg3 Nd5 52. Ree1 Nf4 53. Rc2 Bg8 54. Kh2 Rd8 55. Rd2 Nd5 56.
Kg3 Nf4 57. Re3 Nd5 58. Rf3 Be6 59. Kg2 Nf6 60. Ne3 Rc1 61. Nf5+ Kh7 62.
Nxh6 Bd5 63. Nhf7 Rg8 64. Rd3 Nxh5 65. Kf2 Bxf3 66. Rxf3 Nf4 67. Nxg5+ Rxg5
68. Rxf4 Rg7 69. Ke2 Rgc7 70. Rf3 Kg7 71. Rd3 Kf6 72. Nc4 R1xc4 73. bxc4
Rxc4 74. Rf3+ Ke7 75. Kd3 Rxa4 76. Ke4 b5 77. g5 b4 78. g6 Ra1 79. Rf5 a4
80. g7 Rg1 81. Ra5 a3 82. Kd5 Rxg7 83. Kc4 Rg4 84. Kxb4 Rxd4+ 85. Kxa3 Rd7
86. Kb4 Ke6 87. Ra1 Kf7 88. Rf1+ Ke8 89. Ra1 Rc7 90. Re1+ Kd7 91. Rf1 Kd6
92. Rd1+ Ke5 93. Re1+ Kf6 94. Rf1+ Ke7 95. Re1+ Kf7 96. Rf1+ Kg6 97. Rg1+
Kf5 98. Rf1+ Ke4 99. Re1+ Kd3 100. Rd1+ Ke2 {User Adjudication} *[/pgn]

Komodo 5 1 r2 is rated 3055 on the CEGT blitz list on one thread, almost exactly 500 below Dragon 2. In general at 3000 level or so an average non-edge pawn is worth about 300 elo, and here you also gave the weaker engine White ( so it's "pawn and a move" odds). I would expect Dragon 2 to win a match played that way, perhaps by a score like 7 to 3 or so, with the old engine drawing lots of games but winning few if any. If we use CEGT blitz ratings (because they use Ordo unlike CCRL which uses BayesElo which contracts rating differences), a good guide is that f7 odds is about 400 elo, f2 + c2 about 600, f7 + c7 about 700, knight odds about 1000 elo, and rook odds about 1200 elo, though the exact numbers depend on time control, threads, and specific engine strengths and weaknesses. So to be able to give knight odds in Rapid to Carlsen and break even, we would need a CEGT blitz rating of about 3850 if all of my assumptions are correct. With SF 14 on 8 threads at 3737 this seems doable, but unfortunately I don't think it's linear, the handicap value grows as the level goes up, so even 3850 won't be enough for that. The problem is that beyond a certain point, playing "better" just doesn't help win a clearly lost game.

I am confused are you saying or implying that Carlsen would have a hard time beating Komodo Dragon2 with the c7 + f7 pawn odds at TC 15'+10" ?

[lkaufman]

Komodo Dragon2 Vs Irina at 5'+3" with contempt =205 Based on CCRL rating I believe that if Irina is rated 1435 and Jose rated 1600 and he beat Komodo 13.3 the rating difference between engines of CCRL and human favor human at Queen Odds when comparing CCRL to Humans when we thought that CCRL is equal to 150 rating points less than Humans
[pgn][Event "Queen Odds"]
[Date "2021.09.07"]
[Round "2"]
[White "Dragon-2-64bit-avx2"]
[Black "Irina"]
[Result "1-0"]
[BlackElo "1435"]
[Time "13:11:31"]
[WhiteElo "3550"]
[TimeControl "300+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "91"]
[WhiteType "program"]
[BlackType "program"]

1. e3 Nc6 2. a3 e5 3. d3 d5 4. b3 Nf6 5. Bb2 Bg4 6. Nd2 Nd7 7. h3 Qh4 8. g3
Qh6 9. Ne2 Bxe2 10. Bxe2 Bc5 11. Nf1 Qg6 12. g4 O-O 13. b4 Bd6 14. h4 Nb6
15. h5 Qe6 16. h6 gxh6 17. Ng3 Ne7 18. Nf5 Nxf5 19. gxf5 Qxf5 20. Rh5 Qf6
21. f4 Rae8 22. O-O-O d4 23. Rf1 Qg7 24. Rhh1 Kh8 25. Rhg1 Qf6 26. fxe5 Qe6
27. exd6 Qxe3+ 28. Kb1 c6 29. Bd1 Nd5 30. d7 Re7 31. Bg4 h5 32. Bf5 Rd8 33.
Rh1 Re5 34. Rxh5 Nf6 35. Rhh1 Nxd7 36. Rxh7+ Kg8 37. Rhh1 Nf6 38. Bc1 Qe2
39. Bh6 Rxf5 40. Rxf5 Ng4 41. Rg5+ Kh7 42. Rg7+ Kh8 43. Rxf7 Qg2 44. Rh4
Rb8 45. Bg7+ Kg8 46. Re7 Qf1+ 47. Kb2 Qf7 48. Rxf7 Kxf7 49. Rxg4 Rg8 50.
Rxd4 Rxg7 51. Rd7+ Kf6 52. Rxg7 Kxg7 53. Kb3 Kf6 54. Kc4 Ke5 55. Kc5 Kf4
56. Kd6 Ke3 57. Kc7 Kd2 58. c4 Kxd3 59. c5 Kc4 60. Kxb7 Kb5 61. Kxa7 Ka4
62. Kb6 Kxa3 63. b5 cxb5 64. Kxb5 Kb3 65. c6 Kc3 66. c7 Kd4 67. c8=Q Ke4
68. Qg4+ Kd3 69. Kb4 Ke3 70. Kc3 Kf2 71. Kd3 Kf1 {Black resigns} *[/pgn]

Those are two nice games, especially the rook sacrifice in the first one. On the general topic of simulating humans using engines in handicap games, I'm finding that I can do a pretty good job of simulating a "par" grandmaster in Rapid like this: I think that the CEGT blitz ratings for engines are at about the level such that if an engine on that list played with 2' + 1" and no ponder against a human with 15' + 10" with the same FIDE rating as the CEGT blitz rating of the engine, it would be roughly fair. The engine Benjamin 1.0 by Ed Schroder is the one that has performed best with knight odds compared to other similarly rating engines, so I'm using that. It has a 2530 CEGT blitz rating, so I'm saying that "Benjamin" playing 2' + 1" blitz is a good simulation of a human GM rated 2530 FIDE playing 15' + 10" Rapid. Benjamin beat Dragon 2 (Contempt 100) in 2' + 1" knight odds blitz by 70 elo, so Dragon 2 performed at 2460. In its two serious knight odds matches at 15' + 10" Rapid with titled human players, its performance rating in those 15 games was 2460 FIDE! Perfect agreement! Of course this must be lucky, but it suggests that this is a pretty reasonable way to simulate such matches. There were a lot of details that were different between those tests, but I'm finding that most of these things just don't matter in handicap play.

This is interesting it took Komodo Dragon2 a total of 10 versions to give a pawn odds to Komodo51r2 and be even

[pgn][Event "10th versions later Pawn odds"]
[Site "MININT-UB2PIMJ"]
[Date "2021.09.07"]
[Round "?"]
[White "Komodo51r2-64bit"]
[Black "Dragon-2-64bit-avx2"]
[Result "1/2"]
[BlackElo "3580"]
[Time "15:59:42"]
[WhiteElo "3107"]
[TimeControl "1800+15"]
[SetUp "1"]
[FEN "rnbqkbnr/pp1ppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1"]
[Termination "adjudication"]
[PlyCount "192"]
[WhiteType "program"]
[BlackType "program"]


1. e4 e5 2. Nf3 Nc6 3. Bc4 Bc5 4. c3 Nf6 5. d4 exd4 6. cxd4 Bb4+ 7. Bd2
Bxd2+ 8. Nbxd2 d5 9. exd5 Nxd5 10. Qb3 Nce7 11. Qb5+ Qd7 12. O-O Qxb5 13.
Bxb5+ Kf8 14. Rfe1 g6 15. Ne4 h6 16. Ne5 Kg7 17. Rac1 Rd8 18. h4 Bf5 19.
Bc4 Rac8 20. Nc5 Rc7 21. g4 Bc8 22. f3 b6 23. Ncd3 a5 24. Kf2 g5 25. h5 Be6
26. a3 Rb7 27. b3 Rd6 28. a4 Rc7 29. Bb5 Rb7 30. Ba6 Ra7 31. Nc4 Rd8 32.
Bb5 Rb7 33. Red1 f5 34. Re1 Bg8 35. Nde5 fxg4 36. fxg4 Nf6 37. Red1 Be6 38.
Kf3 Ned5 39. Bc6 Re7 40. Bxd5 Nxd5 41. Kg3 Rc7 42. Rd2 Nf6 43. Re1 Bd5 44.
Rdd1 Be4 45. Re2 Bd5 46. Rc2 Rf8 47. Rcc1 Rfc8 48. Re1 Ne4+ 49. Kh2 Nf6 50.
Re3 Be6 51. Kg3 Nd5 52. Ree1 Nf4 53. Rc2 Bg8 54. Kh2 Rd8 55. Rd2 Nd5 56.
Kg3 Nf4 57. Re3 Nd5 58. Rf3 Be6 59. Kg2 Nf6 60. Ne3 Rc1 61. Nf5+ Kh7 62.
Nxh6 Bd5 63. Nhf7 Rg8 64. Rd3 Nxh5 65. Kf2 Bxf3 66. Rxf3 Nf4 67. Nxg5+ Rxg5
68. Rxf4 Rg7 69. Ke2 Rgc7 70. Rf3 Kg7 71. Rd3 Kf6 72. Nc4 R1xc4 73. bxc4
Rxc4 74. Rf3+ Ke7 75. Kd3 Rxa4 76. Ke4 b5 77. g5 b4 78. g6 Ra1 79. Rf5 a4
80. g7 Rg1 81. Ra5 a3 82. Kd5 Rxg7 83. Kc4 Rg4 84. Kxb4 Rxd4+ 85. Kxa3 Rd7
86. Kb4 Ke6 87. Ra1 Kf7 88. Rf1+ Ke8 89. Ra1 Rc7 90. Re1+ Kd7 91. Rf1 Kd6
92. Rd1+ Ke5 93. Re1+ Kf6 94. Rf1+ Ke7 95. Re1+ Kf7 96. Rf1+ Kg6 97. Rg1+
Kf5 98. Rf1+ Ke4 99. Re1+ Kd3 100. Rd1+ Ke2 {User Adjudication} *[/pgn]

Komodo 5 1 r2 is rated 3055 on the CEGT blitz list on one thread, almost exactly 500 below Dragon 2. In general at 3000 level or so an average non-edge pawn is worth about 300 elo, and here you also gave the weaker engine White ( so it's "pawn and a move" odds). I would expect Dragon 2 to win a match played that way, perhaps by a score like 7 to 3 or so, with the old engine drawing lots of games but winning few if any. If we use CEGT blitz ratings (because they use Ordo unlike CCRL which uses BayesElo which contracts rating differences), a good guide is that f7 odds is about 400 elo, f2 + c2 about 600, f7 + c7 about 700, knight odds about 1000 elo, and rook odds about 1200 elo, though the exact numbers depend on time control, threads, and specific engine strengths and weaknesses. So to be able to give knight odds in Rapid to Carlsen and break even, we would need a CEGT blitz rating of about 3850 if all of my assumptions are correct. With SF 14 on 8 threads at 3737 this seems doable, but unfortunately I don't think it's linear, the handicap value grows as the level goes up, so even 3850 won't be enough for that. The problem is that beyond a certain point, playing "better" just doesn't help win a clearly lost game.

I am confused are you saying or implying that Carlsen would have a hard time beating Komodo Dragon2 with the c7 + f7 pawn odds at TC 15'+10" ?

Yes, that would follow from my math, although I hadn't thought about it. But given the record, this seems right. Nakamura lost 6.5 to 1.5 at that time control with 2 White pawn odds, b2/c2 plus f2/g2. There isn't much difference between b2 and c2 odds or between f2 and g2 odds, so this is a fair statement of how he performed at the equivalent of f2 + c2. Now I think he got upset the second day and played below par, and Carlsen is a bit stronger than Naka in Rapid, but not a lot, so maybe Carlsen would have lost 5.5 to 2.5 or so. The first move is a big deal in these games, enough to swing the expected result to about 4 to 4 at c7 + f7 odds, consistent with my math. Now it is true that f7 odds is a bit larger than g7 even though f2 and g2 odds are similar, so maybe Carlsen would be a favorite at c7 + f7, but not a big favorite. Maybe f7 + c7 is more like 750 elo rather than 700 due to the specific problems Black encounters with f7 missing, but anyway that's a fairly minor point. Of course if the handicap is always the same two pawns he could prepare heavily and learn from each game, so perhaps he would be favored. But if we switch around each game, between the four options (c7 f7, b7 f7, c7 g7, and b7 g7), then I think it would really be a tossup. It also depends on what we do; if we're playing for a million bucks, we'll train Dragon heavily to play those handicaps well!