LC0 40x256 (ID 108) (v0.40.2-rc2) vs LC0 20x256 (ID 42549) (v0.21.2)

[Hai]

The two strongest LC0s:
LC0 40x256 (ID 108) (v0.40.2-rc2) vs LC0 20x256 (ID 42549) (v0.21.2)

The 20 possible opening moves
40 games
Ponder on
LC0 40x256 with 1x RTX 2080 Ti
LC0 20x256 with 1x RTX 2080 Ti
Time control: 1 min per game + 10 seconds per move.

Result:
LC0 40x256 19.5 points
LC0 20x256 20.5 points
+1 =37 -2
Winning percentage = 48.75%

= LC0 40x256 is 9 elo weaker than LC0 20x256.
http://www.mediafire.com/file/eccp53cla ... 9.pgn/file

I'm very happy about this result because a month ago I have tested also the two best 40x256 and 20x256 LC0s and 40x256 was 75 elo weaker. So I avoided to tell that result, because it was not my intention to kill a new project before it starts. Sorry.

Take a look here if you missed some opinions from lkaufman:
viewtopic.php?f=6&t=70772

Congratulations to the 40x256 developers and supporters:)
I hope you can get more training games per day, because you have only 10% of what the main LC0 (20x256) is training per day!
http://157.230.189.191:8080

[Hai]

[pgn][Event "\[207\] LC0 40x256 (ID 108) vs 20x256 (ID "] [Site "Lokaler Computer"] [Date "2019.06.14"] [Round "16"] [White "Lc0 v0.21.2"] [Black "Lc0 v0.40.2-rc2"] [Result "1-0"] [ECO "E08"] [WhiteElo "3600"] [BlackElo "3650"] [PlyCount "353"] [EventDate "2019.??.??"] {[%evp 0,27,27,29,34,6,2,-19,0,10,6,-2,0,8,7,17,27,26,13,9,8,-8,-10,-9,-9,-10, -7,-39,-35,-4]} 1. d4 Nf6 2. c4 e6 3. g3 Bb4+ 4. Bd2 Be7 5. Bg2 d5 6. Nf3 O-O 7. O-O c6 8. Qc2 Nbd7 9. Bf4 b6 10. Rd1 Bb7 11. Ne5 Nh5 12. Bd2 Nhf6 13. cxd5 cxd5 14. Nc6 Qe8 15. Nxe7+ Qxe7 16. Nc3 Rfc8 17. Qb3 Ne4 18. Rac1 Ndf6 19. Bf4 Nh5 20. Bd2 Nhf6 21. Nxe4 dxe4 22. Bg5 h6 23. Bxf6 Qxf6 24. e3 Bd5 25. Qa4 Qd8 26. Bf1 a6 27. Qb4 h5 28. a3 h4 29. Rxc8 Rxc8 30. Qd2 a5 31. Rc1 hxg3 32. hxg3 Rxc1 33. Qxc1 Qd7 34. Qc3 Qc6 35. Be2 Kf8 36. Kf1 Ke7 37. Ke1 Kd7 38. Kd2 f6 39. Bf1 Kd6 40. Be2 Qxc3+ 41. bxc3 Bc6 42. c4 g5 43. Bd1 Kd7 44. Kc3 Kd6 45. Bc2 Kd7 46. Bb3 Kd6 47. Kb2 Bb7 48. Ba4 Kc7 49. Kc3 Kd6 50. Be8 Ke7 51. Bb5 Kd6 52. Ba4 Kc7 53. Kc2 Kd6 54. Kc3 Kc7 55. Kc2 Kd6 56. Kb2 Kc7 57. Kc3 Kd6 58. Be8 Ke7 59. Bb5 Kd6 60. Kc2 Bc8 61. Be8 Bb7 62. Kb3 Bc8 63. Ka4 Ba6 64. c5+ bxc5 65. dxc5+ Kxc5 66. Kxa5 Bc8 67. Ka4 Kc4 68. Bb5+ Kc5 69. Be2 e5 70. Kb3 Be6+ 71. Kc3 Bd7 72. Bc4 f5 73. Bf7 f4 74. Bc4 Bc6 75. Be6 Kb5 76. Kb3 Bb7 77. Bc4+ Ka5 78. Bg8 Kb5 79. Bc4+ Ka5 80. Be6 Kb5 81. Bd7+ Kc5 82. Kc3 Ba6 83. a4 Bb7 84. Bb5 Bd5 85. Ba6 Be6 86. Bb7 Kb6 87. Bxe4 fxe3 88. fxe3 Kc5 89. Bd3 Bd5 90. a5 Bc6 91. a6 Bd5 92. Bf5 Kb6 93. Kb4 Bf3 94. e4 g4 95. Kc4 Kxa6 96. Kd5 Kb7 97. Kxe5 Kc7 98. Ke6 Bxe4 99. Bxe4 Kd8 100. Kf6 Ke8 101. Kg5 Kf7 102. Kxg4 Kg7 103. Bc6 Kg6 104. Kh4 Kh7 105. g4 Kh6 106. Bg2 Kg6 107. Ba8 Kh6 108. Bb7 Kh7 109. g5 Kg6 110. Kg4 Kg7 111. Bc6 Kg6 112. Kh4 Kg7 113. Bf3 Kg6 114. Kg4 Kg7 115. Kf4 Kg6 116. Be4+ Kg7 117. Kg3 Kg8 118. Bd3 Kg7 119. Kh4 Kf7 120. Kh5 Kf8 121. Ba6 Kf7 122. Bc8 Kf8 123. Bb7 Kf7 124. Bg2 Kf8 125. Kg4 Kg7 126. Bf3 Kg6 127. Kh4 Kg7 128. Be2 Kg6 129. Kg4 Kg7 130. Bb5 Kg6 131. Bd3+ Kg7 132. Kf4 Kf7 133. Ke5 Ke8 134. Kf6 Kd7 135. Kg7 Kd6 136. Bb5 Kd5 137. g6 Ke6 138. Be2 Kf5 139. Kh6 Ke5 140. g7 Kd4 141. g8=Q Ke5 142. Qh7 Kd6 143. Kg7 Ke5 144. Qh6 Kd4 145. Qh5 Kc3 146. Qh8 Kd4 147. Qg8 Kc3 148. Qd8 Kb2 149. Kf8 Kc2 150. Ke8 Kc3 151. Kf7 Kc2 152. Qd7 Kb2 153. Qe7 Kc3 154. Qf8 Kd4 155. Qg8 Ke4 156. Qg7 Ke3 157. Qh8 Kd2 158. Qc8 Ke3 159. Qd8 Kf2 160. Kf6 Kxe2 161. Kf5 Ke3 162. Ke5 Kf3 163. Qg8 Ke3 164. Qf7 Kd3 165. Qc7 Ke3 166. Qd7 Kf3 167. Qg7 Ke3 168. Qf8 Kd3 169. Qc8 Ke3 170. Qd7 Kf3 171. Kf5 Ke3 172. Qd8 Kf3 173. Qe8 Kf2 174. Kg4 Kg2 175. Qe2+ Kh1 176. Kg3 Kg1 177. Qe1# 1-0[/pgn]

[pgn][Event "\[207\] LC0 40x256 (ID 108) vs 20x256 (ID "] [Site "Lokaler Computer"] [Date "2019.06.14"] [Round "21"] [White "Lc0 v0.40.2-rc2"] [Black "Lc0 v0.21.2"] [Result "0-1"] [WhiteElo "3650"] [BlackElo "3600"] [PlyCount "174"] [EventDate "2019.??.??"] 1. f3 e5 2. e4 Bc5 3. f4 d5 4. exd5 Nh6 5. Nf3 e4 6. Ng5 O-O 7. Nc3 Re8 8. Be2 Bb6 9. Ngxe4 c6 10. d6 Nd7 11. d3 Nf6 12. Nxf6+ Qxf6 13. Ne4 Qg6 14. Kf1 Bg4 15. Bxg4 Nxg4 16. a4 a5 17. Ra3 f5 18. d7 Re7 19. h3 Ne3+ 20. Bxe3 Bxe3 21. Qf3 Bd4 22. Nd2 Qe6 23. Ra1 Bxb2 24. Rd1 Bc3 25. Nc4 Rd8 26. Kg1 Rdxd7 27. Kh2 Qe2 28. Rdf1 g6 29. Rf2 Qxf3 30. Rxf3 b5 31. Ne5 Bxe5 32. fxe5 Rxe5 33. c3 b4 34. d4 Re2 35. cxb4 axb4 36. Rd1 Ra2 37. Rb3 Rxa4 38. Rc1 Rd6 39. Rc4 Kf7 40. Rcxb4 Rxb4 41. Rxb4 Ke6 42. Kg3 Kd5 43. Kf3 Re6 44. g3 h6 45. g4 fxg4+ 46. hxg4 h5 47. Ra4 hxg4+ 48. Kxg4 Re4+ 49. Kf3 Rxd4 50. Rxd4+ Kxd4 51. Kf2 g5 52. Kg3 c5 53. Kg2 g4 54. Kh2 Kc3 55. Kg3 Kb4 56. Kh2 g3+ 57. Kxg3 Kb3 58. Kf2 c4 59. Ke1 Kc2 60. Ke2 c3 61. Ke1 Kc1 62. Ke2 c2 63. Kd3 Kb1 64. Kd4 c1=Q 65. Kd5 Kc2 66. Kd4 Qd1+ 67. Ke4 Kc3 68. Ke5 Kc4 69. Ke6 Qf1 70. Ke5 Qe1+ 71. Kf4 Kd4 72. Kf5 Ke3 73. Ke5 Qd1 74. Kf5 Kd4 75. Kf6 Qg1 76. Kf5 Ke3 77. Ke5 Qf1 78. Kd5 Qc1 79. Ke5 Qd2 80. Kf5 Kf3 81. Ke5 Qd1 82. Kf5 Qe1 83. Kf6 Kf4 84. Kg6 Ke5 85. Kg5 Qg1+ 86. Kh4 Kf5 87. Kh5 Qh2# 0-1 [/pgn]

[pgn][Event "\[207\] LC0 40x256 (ID 108) vs 20x256 (ID "] [Site "Lokaler Computer"] [Date "2019.06.14"] [Round "28"] [White "Lc0 v0.21.2"] [Black "Lc0 v0.40.2-rc2"] [Result "0-1"] [WhiteElo "3600"] [BlackElo "3650"] [PlyCount "138"] [EventDate "2019.??.??"] {[%evp 0,13,29,-60,-63,-65,-57,-59,-71,-65,-66,-58,-4,-35,-45,-53]} 1. g4 d5 2. Bg2 Nc6 3. c4 d4 4. h3 e5 5. d3 a5 6. Nf3 Nge7 7. Nbd2 Ng6 8. Ne4 h5 9. gxh5 Rxh5 10. h4 Be7 11. Ng3 Rh8 12. h5 Nf4 13. Bxf4 exf4 14. Ne4 f5 15. Ned2 Bf6 16. Ng1 Qd6 17. Rc1 Qc5 18. Kf1 Nd8 19. Nb3 Qb4 20. Nf3 c5 21. Nbd2 Qb6 22. a3 a4 23. Ne1 Ra6 24. Rb1 Qc7 25. Bf3 Kf8 26. b3 axb3 27. Qxb3 Be7 28. Qb2 Rh7 29. Kg2 Rf6 30. Kf1 Rhh6 31. Kg2 Bd7 32. Kg1 Ba4 33. Ng2 Be8 34. Ne1 Bd6 35. a4 g5 36. hxg6 Rfxg6+ 37. Bg2 Rxh1+ 38. Kxh1 Qe7 39. Ndf3 Bc7 40. Qd2 b6 41. Kg1 Bc6 42. Kf1 Nf7 43. Qa2 Nh6 44. Qa3 Ng4 45. Rb5 Qh7 46. Kg1 Nh2 47. Ne5 Bxg2 48. Nxg6+ Qxg6 49. Nxg2 f3 50. exf3 Nxf3+ 51. Kf1 Nh2+ 52. Ke2 Qxg2 53. a5 Ng4 54. Kd1 Qxf2 55. Kc1 Ne3 56. Kb1 Nd1 57. Rb3 Qe1 58. Kc2 Bf4 59. axb6 Qd2+ 60. Kb1 Nc3+ 61. Rxc3 dxc3 62. Qxc5+ Kg7 63. Qe7+ Kg6 64. Qe6+ Kg5 65. Qg8+ Kh4 66. Qh7+ Kg3 67. Qg6+ Kf2 68. Qg2+ Kxg2 69. c5 Qb2# 0-1 [/pgn]

[Laskos]

The two strongest LC0s:
LC0 40x256 (ID 108) (v0.40.2-rc2) vs LC0 20x256 (ID 42549) (v0.21.2)

The 20 possible opening moves
40 games
Ponder on
LC0 40x256 with 1x RTX 2080 Ti
LC0 20x256 with 1x RTX 2080 Ti
Time control: 1 min per game + 10 seconds per move.

Result:
LC0 40x256 19.5 points
LC0 20x256 20.5 points
+1 =37 -2
Winning percentage = 48.75%

= LC0 40x256 is 9 elo weaker than LC0 20x256.
http://www.mediafire.com/file/eccp53cla ... 9.pgn/file

I'm very happy about this result because a month ago I have tested also the two best 40x256 and 20x256 LC0s and 40x256 was 75 elo weaker. So I avoided to tell that result, because it was not my intention to kill a new project before it starts. Sorry.

Take a look here if you missed some opinions from lkaufman:
viewtopic.php?f=6&t=70772

Congratulations to the 40x256 developers and supporters:)
I hope you can get more training games per day, because you have only 10% of what the main LC0 (20x256) is training per day!
http://157.230.189.191:8080

Thanks, impressive!
What is this v0.40.2 engine? Where can one find it?

[lkaufman]

The two strongest LC0s:
LC0 40x256 (ID 108) (v0.40.2-rc2) vs LC0 20x256 (ID 42549) (v0.21.2)

The 20 possible opening moves
40 games
Ponder on
LC0 40x256 with 1x RTX 2080 Ti
LC0 20x256 with 1x RTX 2080 Ti
Time control: 1 min per game + 10 seconds per move.

Result:
LC0 40x256 19.5 points
LC0 20x256 20.5 points
+1 =37 -2
Winning percentage = 48.75%

= LC0 40x256 is 9 elo weaker than LC0 20x256.
http://www.mediafire.com/file/eccp53cla ... 9.pgn/file

I'm very happy about this result because a month ago I have tested also the two best 40x256 and 20x256 LC0s and 40x256 was 75 elo weaker. So I avoided to tell that result, because it was not my intention to kill a new project before it starts. Sorry.

Take a look here if you missed some opinions from lkaufman:
viewtopic.php?f=6&t=70772

Congratulations to the 40x256 developers and supporters:)
I hope you can get more training games per day, because you have only 10% of what the main LC0 (20x256) is training per day!
http://157.230.189.191:8080

Glad to see it's a close competition now. One bit of advice: don't use time controls with small base to increment ratios like this 6 to 1, especially when testing Lc0. As you know, Lc0 sometimes takes ridiculous numbers of moves to win a won game. It is a total waste of testing time to be playing hundreds of moves in a resignable position at almost the same time limit as the first forty moves were played. Even for normal engines, the minimum ratio for efficient testing is 100 to 1 (i prefer 150 to 1 myself), but for Lc0 it is probably at least 300 to 1. You could easily double or triple the number of games per hour with no loss in quality. Maybe five minutes plus one second or ten minutes plus two seconds. Even half a second increment is way more than enough time to win these trivially won positions. I suppose if you are adjudicating aggressively then this isn't so much of an issue, but probably you are not.

[Hai]

The two strongest LC0s:
LC0 40x256 (ID 108) (v0.40.2-rc2) vs LC0 20x256 (ID 42549) (v0.21.2)

The 20 possible opening moves
40 games
Ponder on
LC0 40x256 with 1x RTX 2080 Ti
LC0 20x256 with 1x RTX 2080 Ti
Time control: 1 min per game + 10 seconds per move.

Result:
LC0 40x256 19.5 points
LC0 20x256 20.5 points
+1 =37 -2
Winning percentage = 48.75%

= LC0 40x256 is 9 elo weaker than LC0 20x256.
http://www.mediafire.com/file/eccp53cla ... 9.pgn/file

I'm very happy about this result because a month ago I have tested also the two best 40x256 and 20x256 LC0s and 40x256 was 75 elo weaker. So I avoided to tell that result, because it was not my intention to kill a new project before it starts. Sorry.

Take a look here if you missed some opinions from lkaufman:
viewtopic.php?f=6&t=70772

Congratulations to the 40x256 developers and supporters:)
I hope you can get more training games per day, because you have only 10% of what the main LC0 (20x256) is training per day!
http://157.230.189.191:8080

Thanks, impressive!
What is this v0.40.2 engine? Where can one find it?

After installing this: https://github.com/LeelaChessZero/lc0/releases
Delete that LC0.exe
and download here: http://157.230.189.191:8080 / Notice: Please update engine, that LC0.exe

You can ask questions "Joe MD" here: https://groups.google.com/forum/#!topic ... 5m-2JCdo-A

[Laskos]

The two strongest LC0s:
LC0 40x256 (ID 108) (v0.40.2-rc2) vs LC0 20x256 (ID 42549) (v0.21.2)

The 20 possible opening moves
40 games
Ponder on
LC0 40x256 with 1x RTX 2080 Ti
LC0 20x256 with 1x RTX 2080 Ti
Time control: 1 min per game + 10 seconds per move.

Result:
LC0 40x256 19.5 points
LC0 20x256 20.5 points
+1 =37 -2
Winning percentage = 48.75%

= LC0 40x256 is 9 elo weaker than LC0 20x256.
http://www.mediafire.com/file/eccp53cla ... 9.pgn/file

I'm very happy about this result because a month ago I have tested also the two best 40x256 and 20x256 LC0s and 40x256 was 75 elo weaker. So I avoided to tell that result, because it was not my intention to kill a new project before it starts. Sorry.

Take a look here if you missed some opinions from lkaufman:
viewtopic.php?f=6&t=70772

Congratulations to the 40x256 developers and supporters:)
I hope you can get more training games per day, because you have only 10% of what the main LC0 (20x256) is training per day!
http://157.230.189.191:8080

Thanks, impressive!
What is this v0.40.2 engine? Where can one find it?

After installing this: https://github.com/LeelaChessZero/lc0/releases
Delete that LC0.exe
and download here: http://157.230.189.191:8080 / Notice: Please update engine, that LC0.exe

You can ask questions "Joe MD" here: https://groups.google.com/forum/#!topic ... 5m-2JCdo-A

Thanks, when back home, I will use your pointers. That 40b net can become a monster at long time controls and strong hardware (checked LTC on imo reliable test suites, improves like no other to long times per position).

[Guenther]

The two strongest LC0s:
LC0 40x256 (ID 108) (v0.40.2-rc2) vs LC0 20x256 (ID 42549) (v0.21.2)

The 20 possible opening moves
40 games
Ponder on
LC0 40x256 with 1x RTX 2080 Ti
LC0 20x256 with 1x RTX 2080 Ti
Time control: 1 min per game + 10 seconds per move.

Result:
LC0 40x256 19.5 points
LC0 20x256 20.5 points
+1 =37 -2
Winning percentage = 48.75%

= LC0 40x256 is 9 elo weaker than LC0 20x256.
http://www.mediafire.com/file/eccp53cla ... 9.pgn/file

I'm very happy about this result because a month ago I have tested also the two best 40x256 and 20x256 LC0s and 40x256 was 75 elo weaker. So I avoided to tell that result, because it was not my intention to kill a new project before it starts. Sorry.

Take a look here if you missed some opinions from lkaufman:
viewtopic.php?f=6&t=70772

Congratulations to the 40x256 developers and supporters:)
I hope you can get more training games per day, because you have only 10% of what the main LC0 (20x256) is training per day!
http://157.230.189.191:8080

Thanks, impressive!
What is this v0.40.2 engine? Where can one find it?

After installing this: https://github.com/LeelaChessZero/lc0/releases
Delete that LC0.exe
and download here: http://157.230.189.191:8080 / Notice: Please update engine, that LC0.exe

You can ask questions "Joe MD" here: https://groups.google.com/forum/#!topic ... 5m-2JCdo-A

You seem not to understand that the confusing and bad/idiotic name 0.40.2 or 0.41.2 for the LC0.exe, or whatever, is just a gimmick name?
https://github.com/joeismad/lc0/commits/release/0.21 =>last 7 'commits'

Actually it just adjudicates draws earlier in training games, thus it is completely irrelevant for your games anyway...

[Hai]

The graph is skyrocketing
LC0 40x256
ID 118

graph.jpg

The development breaks through

[Hai]

The two strongest LC0s:
LC0 40x256 (ID 108) (v0.40.2-rc2) vs LC0 20x256 (ID 42549) (v0.21.2)

The 20 possible opening moves
40 games
Ponder on
LC0 40x256 with 1x RTX 2080 Ti
LC0 20x256 with 1x RTX 2080 Ti
Time control: 1 min per game + 10 seconds per move.

Result:
LC0 40x256 19.5 points
LC0 20x256 20.5 points
+1 =37 -2
Winning percentage = 48.75%

= LC0 40x256 is 9 elo weaker than LC0 20x256.
http://www.mediafire.com/file/eccp53cla ... 9.pgn/file

I'm very happy about this result because a month ago I have tested also the two best 40x256 and 20x256 LC0s and 40x256 was 75 elo weaker. So I avoided to tell that result, because it was not my intention to kill a new project before it starts. Sorry.

Take a look here if you missed some opinions from lkaufman:
viewtopic.php?f=6&t=70772

Congratulations to the 40x256 developers and supporters:)
I hope you can get more training games per day, because you have only 10% of what the main LC0 (20x256) is training per day!
http://157.230.189.191:8080

Glad to see it's a close competition now. One bit of advice: don't use time controls with small base to increment ratios like this 6 to 1, especially when testing Lc0. As you know, Lc0 sometimes takes ridiculous numbers of moves to win a won game. It is a total waste of testing time to be playing hundreds of moves in a resignable position at almost the same time limit as the first forty moves were played. Even for normal engines, the minimum ratio for efficient testing is 100 to 1 (i prefer 150 to 1 myself), but for Lc0 it is probably at least 300 to 1. You could easily double or triple the number of games per hour with no loss in quality. Maybe five minutes plus one second or ten minutes plus two seconds. Even half a second increment is way more than enough time to win these trivially won positions. I suppose if you are adjudicating aggressively then this isn't so much of an issue, but probably you are not.

I know the ridiculous numbers of moves to win a won game.
But it's not a total waste of testing time. Of course won endgames can be easily won very fast with low increment but if I would use five minutes plus one second or ten minutes plus two seconds, this would be clearly in favor of 20x256.
For example in the opening or in the middle game when only one or two seconds increment is left, 20x256 would win much more games. 40x256 is at the moment not improved enough, but it still can be improved much more than 20x256.
With 1 second increment it ends 1 vs 8 points and 32 draws, but with 10 seconds increment it ends 1 vs 2 and 37 draws.
So with more and more time, especially with enough time per move, 40x256 becomes stronger and stronger, while 20x256 is already playing with his full strength, because there are no improvements left.

80x1024 could be an interesting test too:)
Or 40x512.
I thing 256 is just to small.

[lkaufman]

The two strongest LC0s:
LC0 40x256 (ID 108) (v0.40.2-rc2) vs LC0 20x256 (ID 42549) (v0.21.2)

The 20 possible opening moves
40 games
Ponder on
LC0 40x256 with 1x RTX 2080 Ti
LC0 20x256 with 1x RTX 2080 Ti
Time control: 1 min per game + 10 seconds per move.

Result:
LC0 40x256 19.5 points
LC0 20x256 20.5 points
+1 =37 -2
Winning percentage = 48.75%

= LC0 40x256 is 9 elo weaker than LC0 20x256.
http://www.mediafire.com/file/eccp53cla ... 9.pgn/file

I'm very happy about this result because a month ago I have tested also the two best 40x256 and 20x256 LC0s and 40x256 was 75 elo weaker. So I avoided to tell that result, because it was not my intention to kill a new project before it starts. Sorry.

Take a look here if you missed some opinions from lkaufman:
viewtopic.php?f=6&t=70772

Congratulations to the 40x256 developers and supporters:)
I hope you can get more training games per day, because you have only 10% of what the main LC0 (20x256) is training per day!
http://157.230.189.191:8080

Glad to see it's a close competition now. One bit of advice: don't use time controls with small base to increment ratios like this 6 to 1, especially when testing Lc0. As you know, Lc0 sometimes takes ridiculous numbers of moves to win a won game. It is a total waste of testing time to be playing hundreds of moves in a resignable position at almost the same time limit as the first forty moves were played. Even for normal engines, the minimum ratio for efficient testing is 100 to 1 (i prefer 150 to 1 myself), but for Lc0 it is probably at least 300 to 1. You could easily double or triple the number of games per hour with no loss in quality. Maybe five minutes plus one second or ten minutes plus two seconds. Even half a second increment is way more than enough time to win these trivially won positions. I suppose if you are adjudicating aggressively then this isn't so much of an issue, but probably you are not.

I know the ridiculous numbers of moves to win a won game.
But it's not a total waste of testing time. Of course won endgames can be easily won very fast with low increment but if I would use five minutes plus one second or ten minutes plus two seconds, this would be clearly in favor of 20x256.
For example in the opening or in the middle game when only one or two seconds increment is left, 20x256 would win much more games. 40x256 is at the moment not improved enough, but it still can be improved much more than 20x256.
With 1 second increment it ends 1 vs 8 points and 32 draws, but with 10 seconds increment it ends 1 vs 2 and 37 draws.
So with more and more time, especially with enough time per move, 40x256 becomes stronger and stronger, while 20x256 is already playing with his full strength, because there are no improvements left.

80x1024 could be an interesting test too:)
Or 40x512.
I thing 256 is just to small.

I agree with you about the network size and about more time helping the larger network. It is obvious to me when analyzing with Lc0 on my 2080 that it rarely changes its analysis much after just a few seconds, and it cannot possibly be true that there is nothing left to discover in most chess positions after ten seconds! But I think it is just wrong to use roughly equal time per move in testing (I know CCRL and CEGT do it for reasons of historical compatibility, but it is terribly wasteful). If you fix the base to increment ratio at say 100 to 1 and just set the time limit to whatever will average the same amount of time per game as 1' + 10", I don't think your results for the larger network will be any worse. Having more time to think when the game is already essentially over is not as useful as having more time to think before that point.