LCZero vs Dedicated machine from 1990s

[Werewolf]

As some of you may know I've been playing the latest online version of LCZero against an old dedicated chess computer, the Master. I think it was better known as the Milano Pro from the 1990s. It's based on an early Frans Morsch program running in a 20 MHz battery powered unit.

http://www.spacious-mind.com/html/milano_pro.html

It's about 2100 elo.

So far each fresh version of LCZero has lost all its games against the Master, 14-0

Today I played the latest ID 55 version (on the website - 2000 rollouts per move) and the Master beat it easily twice. The Master was playing at G20 in all games, but of course, on much slower hardware.


I then installed the ID 55 version on my PC and gave it 30 minutes for the game (Master still playing at G20) and running on my 1060 Nvidia graphics card, allowing around 100,000 rollouts per move. Given this is 5 doublings over the web version, it ought to be at least 250 elo stronger.

Then it played the Master again:

[White "Master"]
[Black "Lczero ID 55"]

[pgn] 1. e4 e6 2. d4 d5 3. Nc3 Nf6 4. Bg5 dxe4 5. Nxe4 Nbd7 6. c3 Be7 7. Qf3 O-O 8. Bd3 c5 9. Nxf6+ Nxf6 10. Bxf6 Bxf6 11. Qh5 g6 12. Qxc5 b6 13. Qc6 Rb8 14. Qf3 Re8 15. Be4 a5 16. Ne2 Ba6 17. Bc6 Rf8 18. O-O Bg7 19. Rfd1 Rc8 20. Nf4 Rb8 21. Re1 Rc8 22. Be4 a4 23. Rad1 Bc4 24. a3 Bb3 25. Rd3 Qd7 26. Rde3 g5 27. Nh5 f5 28. Bb7 Rc7 29. Nxg7 Qxg7 30. Ba6 Rf6 31. Qh5 Re7 32. Bd3 g4 33. Qh4 Qf8 34. Re5 Rd7 35. R5e2 Rg6 36. Re3 Rh6 37. Qg5+ Rg6 38. Qf4 Rf6 39. R3e2 Rg7 40. c4 b5 41. d5 bxc4 42. Bxc4 Re7 43. Bxb3 axb3 44. dxe6 Rg6 1-0 [/pgn]

The Master won a pawn with ease, and despite dithering never looked in trouble.

However, I suspect it's days are limited.

But for now it's 15-0 and counting...

[Leo]

Keep us posted. Its a good measuring stick.

[IanO]

Dedicated oldies are good measuring sticks for weak engines. Maybe get the German forum interested, there is a whole spectrum of even weaker oldies which could be use to measure this initial progress. Check the dedicated Activ list for possibilities.

[Werewolf]

Dedicated oldies are good measuring sticks for weak engines. Maybe get the German forum interested, there is a whole spectrum of even weaker oldies which could be use to measure this initial progress. Check the dedicated Activ list for possibilities.

I have about 10 dedicated machines and using the MESS emulator access to about 30 more.

There are weaker ones i could try, yes. But now I think The Master (Milano Pro) may really be running out of time. It won't keep winning for long.

[Werewolf]

I am now testing the latest ID 78 against the Master.
As usual the Master (Milano Pro, 20 MHz, Frans Morsch) played at G20 minutes.

LCZ ID 78 played from the server on "slow mode", so 2000 rollouts per move.

This is now the 6th match and up to now The Master has won every single game making the score 15 - 0.

How would the new version do?

[Event "1st Match Game (6th match)"]
[Date "2018.04.02"]
[White "Master"]
[Black "LCZ ID 78"]
[Result "1-0"]

[pgn]1. e4 e5 2. Nf3 Nc6 3. Bb5 Nf6 4. O-O Nxe4 5. d4 exd4 6. Re1 f5 7. Nbd2 Be7 8. Nxe4 fxe4 9. Rxe4 O-O 10. c4 Bf6 11. Qa4 Ne7 12. c5 Ng6 13. Nxd4 c6 14. Bc4+ Kh8 15. Qa3 Be7 16. Bd2 a5 17. Rae1 b5 18. Bd3 b4 19. Bxb4 axb4 20. Qxa8 Bxc5 21. Nf3 h6 22. Rc4 Nf4 23. Bf1 Bb6 24. Rc2 Nh3+ 25. gxh3 Rxf3 26. Bg2 Rf8 27. Ree2 Bd4 28. Qb8 Qf6 29. Qxb4 Ba6 30. Red2 c5 31. Qa3 Re8 32. Rc1 Qf4 33. Rcd1 Be2 34. Rxd4 cxd4 35. Rb1 d3 36. Qb3 Rf8 37. Qd5 Qxf2+ 38. Kh1 Rf5 39. Qxd7 Bf3 $4 40. Qc8+ Kh7 41. Qxf5+ Kh8 42. Bxf3 d2 43. Qf8+ Kh7 44. Be4+ g6 45. Qxf2 d1=B 46. Qf7+ Kh8 47. Rxd1 g5 48. Rd8# 1-0 [/pgn]

Some highlights: LCZ ID 78 played a fascinating game with an interesting exchange sac.
The Master was better but then LCZ outplayed it and got a winning position! Unfortunately, it didn't have the tactical firepower to finish the game off and an incredible blunder on move 39 handed the Master the full point. Even so, I was shocked at how much LCZ had caught up.

Score: 16 - 0 to The Master

[Werewolf]

This is the 2nd game in the match.

What started as rather boring opening had some interesting highlights.

First, LCZ declined swapping bishops on move 18, which MAY show it understands the value of the bishop pair.

Secondly, it gave up its "useless" c2 pawn on move 20, which I think was a mistake, because although the pawn was doubled, isolated etc. it was stopping black's pawns. This allowed the Master to create a passed pawn and I assumed it would be another win for the Master. Not so!

LCZ outplayed the Master and scored its first win!

[Event "2nd Match Game (6th match)"]
[Date "2018.04.02"]
[White "LCZ ID 78"]
[Black "Master"]
[Result "1-0"]

[pgn]1. e4 e5 2. Nf3 Nc6 3. Nc3 Nf6 4. d4 exd4 5. Nxd4 Bb4 6. Nxc6 bxc6 7. Bd3 d5 8. O-O O-O 9. exd5 cxd5 10. a3 Bxc3 11. bxc3 Qd6 12. Qf3 Bg4 13. Qg3 Qxg3 14. hxg3 Rfe8 15. f3 Be6 16. Bf4 c5 17. Bb5 Bd7 18. Ba6 Ba4 19. Bb7 Rad8 20. Rfe1 Bxc2 21. Rxe8+ Nxe8 22. Rc1 Bf5 23. g4 Be6 24. a4 d4 25. cxd4 cxd4 26. a5 h6 27. a6 g5 28. Bd2 Nd6 29. Ba5 Re8 30. Bc7 Nb5 31. Ba5 Kg7 32. Bb4 Rd8 33. Ba5 Rd6 34. Bb4 Rd7 35. Kf2 Rd8 36. Be4 Nd6 37. Bd3 Rd7 38. Rc6 Ne8 39. Bc5 Kf6 40. Bf8 Ke5 41. Bxh6 f6 42. Rc5+ Rd5 43. Bf8 Nd6 44. Rc6 Bd7 45. Bxd6+ Rxd6 46. Rc7 Be6 47. Rxa7 Rd8 48. Re7 Kd6 49. Rh7 Ra8 50. Kg3 Bb3 51. a7 Bd5 52. Bb1 Kc6 53. Bd3 Kd6 54. Bf5 Rd8 55. Kf2 Ke5 56. Rd7 Rxd7 57. Bxd7 Kd6 58. Bf5 Kc7 59. Ke2 Bc4+ 60. Bd3 Bd5 61. Be4 Bxe4 62. fxe4 Kb7 63. Kd3 Kxa7 64. Kxd4 Kb6 65. Kd5 Kb5 66. Ke6 Kc5 67. Kxf6 Kd4 68. e5 Ke4 69. e6 Kf4 70. e7 Kxg4 71. e8=Q Kg3 72. Kxg5 Kxg2 73. Qg8 Kf3 74. Kf5 Ke3 75. Ke5 Kd2 76. Kd4 Kd1 77. Kd3 Kc1 78. Qb8 Kd1 79. Qb1# 1-0 [/pgn]

Score: 16 - 1 to the Master.

We need more games to establish if the ID 78 is as good as the Master at 2000 rollouts per move, but it must be close now. If I installed this version on my PC and allowed 100,000 rollouts per move (as I did with ID 55) I think LCZ would win nearly all of its games.

Exciting times.

Perhaps the Tasc R30 needs to be woken from its slumber...

[Werewolf]

There may still be a way to go. The following game was played on my local PC with over 100,000 rollouts per move.


[Date "2018.04.02"]
[White "Master"]
[Black "LCZero ID78" on Nvidia 1060 GPU]
[Result "1-0"]


[pgn] 1. e4 e6 2. d4 d5 3. Nd2 c5 4. exd5 exd5 5. Ngf3 cxd4 6. Bb5+ Nc6 7. O-O Bd6 8.
Nb3 Nge7 9. Nfxd4 O-O 10. Re1 Nxd4 11. Nxd4 Ng6 12. Qh5 Bc5 13. c3 Qb6 14. Be2
Bxd4 15. cxd4 Qxd4 16. Rd1 Qb4 17. Qxd5 Re8 18. Qb5 Qe7 19. Be3 h6 20. Rac1 Nh4
21. Rc7 Qe4 22. Bf1 Nf5 23. Bd3 a6 24. Bxe4 axb5 25. Rxc8 Raxc8 26. Bxf5 Rcd8
27. Rxd8 Rxd8 28. Kf1 Rd1+ 29. Ke2 Rg1 30. h3 b4 31. Be4 b5 32. Kd2 Ra1 33. Bd5
Kf8 34. Bc5+ Ke8 35. Bxb4 g5 36. a3 Rb1 37. Bc6+ Kd8 38. Bc3 Kc7 39. Bxb5 [/pgn]
1-0

[Werewolf]

Well there are ups and downs. As LCZ moves into striking distance of the strong dedicated chess computers from the 1990s it loses some games and wins others.

On my PC I was able to try ID 80 at 100,000 rollouts per move, and it CRUSHED the Master in 14 moves as black! I played the game out just to be sure, but there was no need.

[Date "2018.04.03"]
[White "Master"]
[Black "LCZ ID80 Nvidia 1060, 100,000 rollouts per go"]
[Result "0-1"]

[pgn]1. e4 e6 2. d4 d5 3. Nc3 h6 4. Bf4 Bb4 5. exd5 exd5 6. Qe2+ Ne7 7. Qe5 O-O 8.
Qxc7 Qe8 9. O-O-O Nbc6 10. Na4 Ba5 11. Qd6 Bf5 12. Nf3 Bb4 13. Nc5 Rd8 14. Qc7
Bxc5 15. Qxb7 Bb6 16. h3 Bc8 17. Qa8 Qd7 18. Bb5 Bb7 19. Ne5 Nxe5 20. Bxd7 Bxa8
21. Ba4 N5g6 22. Bh2 Rc8 23. c3 Bc6 24. Bc2 f6 25. Rhe1 Rfd8 26. Rd3 h5 27.
Rde3 Kf7 28. Kd2 Bc7 29. Bxc7 Rxc7 30. g3 h4 31. b3 a6 32. a3 Rh8 33. Rg1 Bd7
34. g4 Nf4 35. Rge1 Bc8 36. Rf3 Ne6 37. Rfe3 Nf4 38. Rf3 g5 39. Rh1 Nxh3 40.
Rfxh3 Bxg4 41. Re3 Nf5 42. Ree1 Nd6 43. Bd3 Bf3 44. Rh2 Ne4+ 45. Bxe4 Bxe4 46.
c4 Bf3 47. Kc3 Rh7 48. Re3 g4 49. Kb4 Be4 50. Ka5 f5 51. cxd5 h3 52. Kxa6 Bxd5
53. Kb6 Rc2 54. Re5 Be4 55. Rxe4 fxe4 56. Rh1 e3 57. fxe3 g3 58. Rf1+ Kg6 59.
Rf3 g2 60. Rg3+ Kf7 61. e4 h2 62. Rxg2 Rxg2 63. Kb5 h1=Q 64. e5 Rhg7 65. d5 Rd2
66. Kb6 Qxd5 67. b4 Qe6+ 68. Ka5 Qc8 69. b5 Rb2 70. Kb6 Ke6 71. a4 Rb7+ 72. Ka5
Qa8# [/pgn]
0-1


wooo..!

[Werewolf]

The weaker web version - currently ID 78 - only has 2000 rollouts per go so it's not as strong. Here's the Vancouver 68000 flexing its muscles:

[Date "2018.04.03"]
[White "Vancouver 68000"]
[Black "LCZ ID 78"]
[Result "1-0"]

[pgn] 1. e4 e5 2. Nf3 Nc6
3. Bc4 Nf6 4. d3 Bc5 5. Nc3 d6 6. Be3 Bxe3 7. fxe3 Na5 8. Bb5+ c6 9. Ba4 b5 10.
Bb3 Nxb3 11. axb3 b4 12. Ne2 a5 13. O-O Ng4 14. Qd2 O-O 15. h3 Nf6 16. Ng3 d5
17. Nxe5 dxe4 18. d4 c5 19. Rf2 cxd4 20. exd4 h5 21. c4 Ra6 22. Re1 Re8 23. Qe3
a4 24. bxa4 h4 25. Ne2 Rxa4 26. Qg5 Nh7 27. Qh5 Rf8 28. Rxf7 Nf6 29. Rxf8+ Qxf8
30. Qxh4 Qd6 31. Ng6 Bf5 32. c5 Qe6 33. Nef4 Qe8 34. Qh8+ Kf7 35. Ne5+ Ke7 36.
Qxg7+ Kd8 37. Qxf6+ Kc7 38. Nd5+ Kc8 39. Qd6 Ra7 40. Nb6+ Kb7 41. c6+ Kxb6 42.
Nc4+ Ka6 43. Ra1+ Kb5 44. Qc5# [/pgn]
1-0


The Vancouver 68000 is this machine:

https://www.schach-computer.info/wiki/i ... uver_68000

It's about the same strength as the Master, slightly less strong tactically and much more conservative in playing style. However, like other Lang programs it's positional understanding is sound and its endgame is far better than the Master's.

[Werewolf]

I decided to see if the Vancouver 68000 could win again, but this time against "Big Brother", LCZero ID 8 at 100,000 rollouts per move, on my PC.

This happened:

[Date "2018.04.03"]
[White "Vancouver 68000"]
[Black "LCZ ID 80 Nvidia 1060 100K rollouts per go"]
[Result "0-1"]

[pgn]1. e4 e6 2. d4 d5 3. Nc3 h6 4. e5 c5 5. Nf3 Nc6 6. Ne2 cxd4 7. Nexd4 Nge7 8.
Bb5 g6 9. c4 dxc4 10. Qa4 Bd7 11. Qxc4 Bg7 12. Qc5 Qb6 13. Qxb6 axb6 14. Nxc6
Nxc6 15. Bf4 g5 16. Bg3 g4 17. Bxc6 Bxc6 18. Nh4 Bf8 19. a3 Bb4+ 20. Kf1 Bc5
21. Rd1 Ba4 22. Rd2 Bb5+ 23. Kg1 Bc6 24. h3 h5 25. Bf4 Rg8 26. Bg3 Rf8 27. hxg4
hxg4 28. Bf4 Rg8 29. Kh2 Be7 30. Kg3 Ra5 31. Rdd1 Bd8 32. Rd3 Ra4 33. Rhd1 Ra8
34. Rh1 Rh8 35. Rdd1 Rg8 36. Rd3 Rh8 37. Rdd1 Ra4 38. Nf5 Rxh1 39. Nd6+ Kf8 40.
Rxh1 Kg7 41. Kxg4 Bxg2 42. Rg1 Bc6 43. Kh3+ Kh7 44. Be3 f5 45. f4 b5 46. Bc1
Bb6 47. Rg3 Bf2 48. Rg5 Rd4 49. Rh5+ Kg7 50. Rg5+ Kf8 51. Rg2 Bxg2+ 52. Kxg2
Rd1 53. Kxf2 Rxc1 54. Ke3 Ke7 55. Nxb5 Rb1 56. b4 b6 57. Nd4 Kd7 58. Kd3 Rf1
59. Ke3 Ra1 60. Nc2 Rh1 61. Nd4 Rh3+ 62. Nf3 Kc6 63. a4 Kd5 64. a5 bxa5 65.
bxa5 Rh8 66. Nd4 Rh3+ 67. Nf3 Rh1 68. Nd4 Re1+ 69. Kf2 Rc1 70. Ne2 Ra1 71. Nc3+
Kc4 72. Ne2 Rxa5 73. Kf3 Ra3+ 74. Kf2 Kc5 75. Kg2 Kd5 76. Kf1 Ke4 77. Kf2 Ra2
78. Kf1 Ke3 79. Nc3 Ra1+ 80. Kg2 Kxf4 81. Kf2 Kxe5 82. Ne2 Ra3 83. Ng3 f4 84.
Ne2 Ke4 85. Kg1 Ke3 86. Nxf4 Kxf4 87. Kf2 Re3 88. Kg2 Re2+ 89. Kf1 Kf3 90. Kg1
Re1+ 91. Kh2 e5 92. Kh3 Rh1# [/pgn]
0-1

Incredible.
Notice the same line as against the Master with the terrible 3...h6?
The Master tried to punish this with tactics, won a pawn but lost a piece. The Vancouver played better, typically sound if uninspired play. But as the game went on, slowly slowly, LCZ turned it around.

I will give these two class machines (The Master & Vancouver 68000) one more chance against an updated LCZ. If they lose again I propose bigger guns:

- Saitek RISC 2500
- Mephisto RISC 1 MB
- London 68030
- Tasc R30

The 90's machines won't go down without a fight.