Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / Dragon 2 by Komodo
- - By MrKris (***) Date 2021-05-05 23:38
Authors' announcement see http://talkchess.com/forum3/viewtopic.php?f=2&t=77244
... Multi-PV search...is sufficiently improved that we no longer advocate using MCTS mode ...
-{2}- is a slight upgrade from the version that beat Stockfish in the fifty game final match of the TCEC chess960 (FRC) championship. -{but see below}-
We believe that the MultiPV search is now more effective than the Stockfish MultiPV search, and that if MultiPV is set to more than 4 ...
-{no data}-


CCRL screenshots from links at:
http://talkchess.com/forum3/viewtopic.php?f=6&t=77252
Parent - - By sarona (*) Date 2021-05-07 00:11
K-Dragon2 is using the same net that K-Dragon 1 came with.  The multi-PV search speed has definitely improved on my 2950X.

On an unrelated note, I have been having a bit (actually, a lot) of trouble compiling the Lc0 binary with Ubuntu. Will keep you posted. I honestly have not spent much time on it over the last few days.

Thanks (again) for the game/tournament/ALL threads. Currently, the COVID situation is very bad in Alberta. There may be curfews in Calgary. The wife and I are vaccinated (Sara, only one dose right now), but my daughters are not (too young).  I am very concerned for their safety. The chess forums/ posters (especially, Rybka, Forchess and Talkchess) take my mind off of that.

Thanks again to all.
Parent - By MrKris (***) Date 2021-05-07 04:26

> The multi-PV search speed has definitely improved on my 2950X.


Interesting! LK did seem quite supportive of it at TalkChess.

> trouble compiling the Lc0 binary


You can send me a message about it if you like. (Mine 21.04 seems okay still.)
I have Lc0 a little weaker with Sf 2700X (16ths) vs. RTX 2060 even after the Cuda 11.2's 32% (on mine) Lc0 nps speedup.

Hope that you and your family can stay safe!!
- By MrKris (***) Date 2021-05-08 00:17
See https://www.sp-cc.de/ for the full list, now with Dragon 2 by Komodo.
- - By MrKris (***) Date 2021-05-09 20:16
http://www.fastgm.de/60-0.60.html
Screenshot of:
Complete list: May 08, 2021
Parent - By MrKris (***) Date 2021-05-28 17:10
Text from above screenshot:
http://fastgm.de/60-0.60.html
Ordo v1.2.6: -a 2993 -A "Critter 1.6a 64-bit"
Complete list: May 27, 2021

Rank    Name            Rating  +-      Games
...
21   Komodo 13.3   3352   6   28250
...
36   Komodo 14.1   3311   4   35750


I have 13.3 so I  gave it 8 treads vs. the latest Stockfish with only 1 thread:

R7 2700X (16th SMT on) Sy.6
G/30 secs. + .5 secs.
Nooman 3-move Testsuite
   Stockfish_21052701_1-thread 128MBh
   Komodo-13.3_8-threads       1GBh 256MBtm
 
300 of 300 games finished:
Score of Stockfish_21052701_1-th vs Komodo-13.3_8-threads: +146 -15 =139 [71.8%]
                                                           ^^^^ ^^^ ^^^^  ^^^^^
...      Stockfish_21052701_1-th playing White: 97 - 5 - 48  [0.807] 150
...      Stockfish_21052701_1-th playing Black: 49 - 10 - 91  [0.630] 150
...      White vs Black: 107 - 54 - 139  [0.588] 300
Elo difference: +162.6  +/-28.6, LOS: 100.0 %, DrawRatio: 46.3 %
               ^^^^^^^
 

Round 124 (game #247 & #248): Stockfish-1th won with both colors. Komodo-8ths loss with white was the shortest win.
[Event "Sf 1-to-8 threads challenge"]
[Site "G/30''+0.5'' Sy.6 Nooman3mv"]
[Date "2021.05.28"]
[Round "124"]
[White "Stockfish_21052701_1-th"]
[Black "Komodo-13.3_8-threads"]
[Result "1-0"]
[ECO "A04"]
[Opening "Reti Opening"]
[Variation "1...c5 2.g3 Nc6 3.Bg2"]
[TimeControl "30+6"]
[Termination "adjudication"]
[PlyCount "140"]
[WhiteType "human"]
[BlackType "human"]

1. Nf3 {book} c5 {book} 2. g3 {book} Nc6 {book} 3. Bg2 {book} e5 {book} 4.
c4 {+0.30/22 0 1.3s} Nf6 {-0.23/22 0 1.6s}

( 4... g6 {Komodo13.3-8th 0-1 Stockfish-1th} 5. d3 Nge7 6. Nc3 Bg7 7. h4 h6 8.
Bd2 d6 9. Qc1 Be6 10. O-O Rc8 11. Re1 f5 12. Rb1 b6 13. b3 f4 14. gxf4 Nf5
15. h5 O-O 16. hxg6 Nh4 17. Kf1 Nd4 18. Ne4 exf4 19. Bc3 d5 20. cxd5 Bxd5
21. Nxh4 Qxh4 22. Nf6+ Bxf6 23. Bxd5+ Kh8 24. Bf3 Rce8 25. Bg2 Rg8 26. e3
f3 27. Bxf3 {0-1} )

5. Nc3 {+0.80/21 0 0.45s} d6
{-0.26/24 0 3.1s} 6. O-O {+0.67/25 0 4.9s} Be7 {-0.22/24 0 1.3s} 7. b3
{+0.66/19 0 0.41s} h6 {-0.28/23 0 1.3s} 8. Bb2 {+0.76/21 0 1.5s} O-O
{-0.24/22 0 0.69s} 9. e3 {+0.70/18 0 0.40s} Bf5 {-0.44/22 0 3.6s} 10. Ne1
{+0.85/21 0 1.2s} Qd7 {-0.13/20 0 1.3s} 11. e4 {+0.74/21 0 0.91s} Be6
{0.0023 1.2s} 12. Nc2 {+0.86/20 0 0.54s} Nd4 {-0.12/21 0 1.4s} 13. Ne3
{+0.81/21 0 0.53s} Qd8 {-0.19/22 0 4.3s} 14. d3 {+0.79/24 0 1.6s} a6
{-0.07/22 0 4.2s} 15. f4 {+0.74/20 0 0.48s} Rb8 {0.0021 1.7s} 16. Kh1
{+0.91/20 0 0.73s} b5 {-0.02/20 0 1.0s} 17. Rf2 {+0.87/21 0 1.4s} Re8
{0.0020 2.1s} 18. f5 {+0.67/23 0 4.1s} Bc8 {+0.32/22 0 1.3s} 19. h4
{+1.03/20 0 1.1s} Bb7 {+0.29/18 0 0.65s} 20. Bc1 {+0.99/20 0 2.4s} Bc6
{+0.37/22 0 2.3s} 21. Bh3 {+0.88/19 0 0.95s} bxc4 {+0.56/19 0 0.51s} 22.
bxc4 {+0.98/18 0 0.43s} a5 {+0.33/20 0 1.3s} 23. Ncd5 {+0.93/21 0 1.7s} a4
{+0.07/19 0 1.1s} 24. Rh2 {+0.84/19 0 0.80s} Nxd5 {+0.42/20 0 1.1s} 25.
cxd5 {+1.06/20 0 0.74s} Bb5 {+0.25/22 0 1.1s} 26. Bf1 {+1.10/22 0 1.1s} Rb7
{+0.16/22 0 2.4s} 27. Ng4 {+0.54/24 0 9.5s} Kh8 {+0.21/20 0 0.74s} 28. Nxh6
{+0.48/21 0 2.6s} gxh6 {+0.71/18 0 0.32s} 29. Bxh6 {+0.74/18 0 0.20s} Bf8
{+0.39/22 0 1.2s} 30. Be3 {-0.07/22 0 1.2s} Kg8 {+0.28/19 0 0.53s} 31. g4
{+0.62/18 0 0.51s} Qd7 {+0.37/20 0 0.50s} 32. f6 {+2.70/16 0 0.19s} Ba6
{0.0025 0.38s} 33. Rg2 {+1.59/21 0 0.76s} Qc8 {-0.06/23 0 0.61s} 34. Rg3
{+3.36/17 0 0.13s} Bb5 {-1.45/20 0 0.50s} 35. Rc1 {+3.75/19 0 0.46s} a3
{-1.62/20 0 0.50s} 36. Bd2 {+4.37/18 0 0.48s} Bd7 {-1.36/21 0 0.50s} 37. g5
{+4.05/18 0 0.24s} Bf5 {-1.78/21 0 0.48s} 38. Bg2 {+4.50/17 0 0.15s} Bh7
{-1.61/20 0 0.38s} 39. h5 {+3.50/21 0 1.1s} Qd8 {-2.51/21 0 0.64s} 40. Qg4
{+3.84/17 0 0.14s} Kh8 {-2.86/21 0 0.50s} 41. g6 {+3.55/18 0 0.16s} fxg6
{-1.78/23 0 0.36s} 42. Rf1 {+3.59/18 0 0.18s} Qd7 {-2.36/22 0 0.48s} 43.
Qxd7 {+3.08/22 0 0.48s} Rxd7 {-2.20/24 0 0.28s} 44. Bh3 {+2.32/21 0 0.64s}
Nf5 {-2.01/23 0 0.34s} 45. Rgg1 {+2.41/20 0 0.22s} gxh5 {-2.13/23 0 0.30s}
46. Bxf5 {+3.02/23 0 0.27s} Bxf5 {-1.53/22 0 0.41s} 47. Rxf5 {+2.99/23 0
0.45s} Rf7 {-1.43/22 0 0.36s} 48. Kh2 {+3.63/23 0 0.32s} Rb8 {-1.63/23 0
0.45s} 49. Bc1 {+3.80/22 0 0.30s} Rb1 {-2.34/25 0 1.1s} 50. Rxh5+ {+3.90/22
0 0.25s} Rh7 {-2.31/24 0 0.31s} 51. Rxh7+ {+3.96/23 0 0.33s} Kxh7 {-2.31/5
0 0s} 52. Bxa3 {+3.94/24 0 0.68s} Rb7 {-2.16/26 0 0.66s} 53. Bc1 {+3.63/27
0 0.53s} Ra7 {-2.33/26 0 0.81s} 54. a3 {+4.48/24 0 0.41s} Rf7 {-2.22/24 0
0.45s} 55. Kg2 {+4.33/22 0 0.51s} Rxf6 {-2.72/21 0 0.34s} 56. a4 {+4.90/21
0 0.30s} Be7 {-2.99/25 0 0.69s} 57. Rf1 {+5.22/22 0 0.37s} Rg6+ {-3.14/23 0
0.40s} 58. Kh3 {+5.20/22 0 0.41s} Rg7 {-3.45/27 0 1.3s} 59. Bd2 {+5.24/23 0
0.45s} Bd8 {-3.65/25 0 0.54s} 60. Rf8 {+5.53/21 0 0.41s} Rd7 {-3.66/25 0
0.50s} 61. Kg4 {+6.12/19 0 0.33s} Kg7 {-3.98/23 0 0.50s} 62. Re8 {+6.23/22
0 0.34s} Kf7 {-4.12/22 0 0.36s} 63. Re6 {+6.76/22 0 0.58s} Bf6 {-4.37/24 0
0.64s} 64. Kf5 {+7.34/21 0 0.39s} Bg7 {-5.36/24 0 0.51s} 65. a5 {+7.91/23 0
0.44s} Bf8 {-6.11/24 0 0.43s} 66. Bh6 {+8.52/22 0 0.46s} Bxh6 {-6.27/27 0
0.57s} 67. Rxh6 {+8.83/23 0 0.52s} Kf8 {-6.88/23 0 0.50s} 68. a6 {+9.01/22
0 0.44s} Kg8 {-6.83/22 0 0.50s} 69. Rf6 {+9.23/21 0 0.41s} Kg7 {-7.44/22 0
0.50s} 70. a7 {+9.31/21 0 0.37s} Rxa7 {-11.26/23 0 0.50s, White wins by
adjudication} 1-0
 
Attachment: Sf-1th_K133-8ths.pgn - Games (831k)
- - By MrKris (***) Date 2021-05-10 19:42
https://www.sp-cc.de/index.htm
KomodoDragon 2.0 MCtS screenshot below:
Parent - - By Peter Grayson (****) Date 2021-05-11 13:08
If the SP-CC games conditions are run with single thread per engine then the Komodo Dragon 2 results cannot reflect LK's predicted 50 Elo improvement for multi-thread efficiencies over Komodo Dragon 1?

Peter
Parent - By MrKris (***) Date 2021-05-11 16:04 Upvotes 1

> If the SP-CC games conditions are run with single thread per engine then the Komodo Dragon 2 results cannot reflect LK's predicted 50 Elo improvement for multi-thread efficiencies over Komodo Dragon 1?


- Yes, what is really needed is a match Stockfish PV=6 vs. KomodoDragon2 PV=6, any PV above 4 as LK below, although no GUI's show the additional PV's in their .pgn they will run it without any problem.

It do not think the SP-CC single thread would hurt anything, but high PV's would reduce the depths for both, lowering quality, so more threads and/or longer time control would be an idea.

sarona, above this thread, wrote: "K-Dragon2 is using the same net that K-Dragon 1 came with.  The multi-PV search speed has definitely improved on my 2950X. ..."
With KD2 only a few Elo above KD1 but KD2 MCTS 84 Elo above KD1 I am speculating that the Mult-PV improvement helped MCTS also.

- However, there is a big 'however' about that:
With Sf about 80 Elo better than KD2 what would be the point of "better multi-PV" when KD2's best is so much worse than Sf's and then the rest of KD2's PV's go down from there?

LK from http://talkchess.com/forum3/viewtopic.php?f=2&t=77244 :
"... We believe that the MultiPV search is now more effective than the Stockfish MultiPV search, and that if MultiPV is set to more than 4 Dragon 2 is now the favorite against Stockfish 13 on typical modern hardware. ..."
If that is technically true --not just subjectively as in 'more human' etc.-- then LK could have easily provided at least one or a few games with roughly favorable high-PV vs. high-PV KD2 vs. Sf (since GUI's will do that, as above, and left it to others to do more games for a better estimate) - but he provided none, not even a single example position.
- By MrKris (***) Date 2021-05-13 03:45
http://www.fastgm.de/16-60-0.6.html
May 12, 2021
Ordo v1.2.6: Fire 7.1 3200
New: 20. Dragon by Komodo Chess 2.0

16 Cores vs GeForce RTX 2070

#   Engine   Elo   Error   Games   Points   (%)   W   D   L   D(%)   CFS
1   Fat Fritz 2 (in SF)   3543   7   3750   2131.00   56.83   603   3056   91   81.5   79
2   Stockfish 210131   3539   8   2500   1363.50   54.54   289   2149   62   86.0   70
...
18   Lc0 0.26.3 RC2 384x30-J92-130   3502   8   4000   3051.50   76.29   2190   1723   87   43.1   57
19   Dragon by Komodo Chess 1.0   3501   5   5750   3006.50   52.29   690   4633   427   80.6   68
20   Dragon by Komodo Chess 2.0   3499   8   2500   1213.50   48.54   189   2049   262   82.0   74
...
- By MrKris (***) Date 2021-05-16 02:45 Edited 2021-05-16 02:50
CCRL Blitz - 15 may 2021
https://ccrl.chessdom.com/ccrl/404/

Note: only 1 KD2 win vs. Sf in 100 games , and only 2 in 100 games vs. FF2 ?!

 
- - By MrKris (***) Date 2021-05-17 01:48 Edited 2021-05-17 01:55
http://www.fastgm.de/10min.html

Time control:  10 minutes / game + 6 seconds / move

NOTE: http://www.fastgm.de/h2h10.html
Head to head statistics vs. Sf13:
KD2 +3 -53 in 300 games; KD1 +2 -57 in 300; KD2mcts +0 -85 in 300; K14.1 + 0 -157 in 300.

1) Stockfish 13                    3493 :   8400 (+5003,=3372,-25),  79.6 %

     vs.                                   :  games (    +,    =,  -),   (%) :   Diff,   SD, CFS (%)
...
     Dragon by Komodo Chess 2.0            :    300 (   53,  244,  3),  58.3 :    +48,    5,  100.0
     Dragon by Komodo Chess 1.0            :    300 (   57,  241,  2),  59.2 :    +57,    4,  100.0
     Dragon by Komodo Chess 2.0 MCTS       :    300 (   85,  215,  0),  64.2 :   +101,    5,  100.0
     Komodo 14.1                           :    300 (  157,  143,  0),  76.2 :   +176,    4,  100.0
 

Also, Komodo 14.1 came in 25 Elo below Stockfish 9 from the Stone Age.
 
Parent - By jpqy (**) Date 2021-05-18 07:31 Upvotes 1
See also results at Ipmanchess : http://ipmanchess.yolasite.com/i9-7980xe.php
- By MrKris (***) Date 2021-05-18 21:17
http://ipmanchess.yolasite.com/i9-7980xe.php

(regular +8 Elo, MCTS +63)
 
- By MrKris (***) Date 2021-05-19 03:49
http://fastgm.de/16-60-0.6.html  -  KD2 MCTS
Time control:  60 seconds / game + 0.6 seconds / move
16 Cores vs GeForce RTX 2070
 
- By MrKris (***) Date 2021-05-24 02:04
CEGT 40/4   (1CPU so far)

http://www.cegt.net/40_4_Ratinglist/40_4_AllVersion/rangliste.html
Not shown here:
KD2.0 1cpu + 12 Elo KD1.0
KD2.0mcts 1cpu + 99 Elo KD1.0mcts 1cpu

Screenshots: http://www.cegt.net/40_4_Ratinglist/40_4_single/rangliste.html
 
Up Topic The Rybka Lounge / Computer Chess / Dragon 2 by Komodo

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill