Mayhem NNUE - New NN engine

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

JohnW
Posts: 381
Joined: Thu Nov 22, 2012 12:20 am
Location: New Hampshire

Re: Mayhem NNUE - New NN engine

Post by JohnW »

Any *.exe's available to download?
User avatar
Sylwy
Posts: 4466
Joined: Fri Apr 21, 2006 4:19 pm
Location: IASI - the historical capital of MOLDOVA
Full name: SilvianR

Re: Mayhem NNUE - New NN engine

Post by Sylwy »

JohnWoe wrote: Tue Dec 01, 2020 10:07 pm
mclane wrote: Tue Dec 01, 2020 10:00 pm Any idea about the ELO ?
Mr. Sylwy tested v0.50 and got around 2790 Elo : http://talkchess.com/forum3/viewtopic.php?f=6&t=75796

I have added opening books and lots improvements. So my estimate is somewhere in 2800 - 3000 Elo.
I have been trying to close the gap against Hakkapeliitta 3.0. I think I'm close but not quite there yet.
Well, after 15 games (in my 100 games match-test) Mayhem 2.5-x64-NNUE (modern version) seems equal to Rhetoric 1.4.3 64-bit (2806 Elo points in CCRL Blitz rating list). Both performance.bin & nn-62ef826d1a6d.nnue SF net are used. TC=4'+2" (12-15 minutes per game).

Image
JohnWoe
Posts: 491
Joined: Sat Mar 02, 2013 11:31 pm

Re: Mayhem NNUE - New NN engine

Post by JohnWoe »

JohnW wrote: Thu Dec 31, 2020 7:51 pm Any *.exe's available to download?
I only posted source code as my compiles are a bit slow on this AMD machine. Missing CPU instructions resulting ~50% speed loss.
Sylwy wrote: Fri Jan 01, 2021 3:37 pm
JohnWoe wrote: Tue Dec 01, 2020 10:07 pm
mclane wrote: Tue Dec 01, 2020 10:00 pm Any idea about the ELO ?
Mr. Sylwy tested v0.50 and got around 2790 Elo : http://talkchess.com/forum3/viewtopic.php?f=6&t=75796

I have added opening books and lots improvements. So my estimate is somewhere in 2800 - 3000 Elo.
I have been trying to close the gap against Hakkapeliitta 3.0. I think I'm close but not quite there yet.
Well, after 15 games (in my 100 games match-test) Mayhem 2.5-x64-NNUE (modern version) seems equal to Rhetoric 1.4.3 64-bit (2806 Elo points in CCRL Blitz rating list). Both performance.bin & nn-62ef826d1a6d.nnue SF net are used. TC=4'+2" (12-15 minutes per game).

Image
Thanks!
Good to know. 2.5 is good version. 2800 Elo is my estimation too. According test games vs Fruit/Glaurung.
Tho my tests are always a bit biased towards Mayhem.
JohnWoe
Posts: 491
Joined: Sat Mar 02, 2013 11:31 pm

Re: Mayhem NNUE - New NN engine

Post by JohnWoe »

Mayhem 2.7 released.

Should be much much stronger than any other version.
Now I test without adjudication.
Thanks to Mr. Sylwy's tournament games!
Even badly handicapped Mayhem is 70% / 150 Elo stronger than Fruit 2.1.

Mayhem 2.7 crushes Hakkapeliitta 3.0 on this Linux box even while badly handicapped. (Doesn't have all NNUE flags. Speed is 30% of optimum.)
There is no way Hakkapeliitta is stronger than Mayhem 2.7.

Mayhem 2.7:
- Contains lots of EGTB knowledge. The most important 7-6-5 men.
- I created EGTB generator in Python which I'm not publishing ever. It's for Mayhem only. I will generate more knowledge for v2.8
- Lots of bugs were fixed.
- Speedups
- Now RAM is used optimally.

Release: https://github.com/SamuraiDangyo/mayhem ... s/tag/v2.7
User avatar
Graham Banks
Posts: 41423
Joined: Sun Feb 26, 2006 10:52 am
Location: Auckland, NZ

Re: Mayhem NNUE - New NN engine

Post by Graham Banks »

JohnWoe wrote: Sat Jan 09, 2021 5:43 pm Mayhem 2.7 released.

Should be much much stronger than any other version.
Now I test without adjudication.
Thanks to Mr. Sylwy's tournament games!
Even badly handicapped Mayhem is 70% / 150 Elo stronger than Fruit 2.1.

Mayhem 2.7 crushes Hakkapeliitta 3.0 on this Linux box even while badly handicapped. (Doesn't have all NNUE flags. Speed is 30% of optimum.)
There is no way Hakkapeliitta is stronger than Mayhem 2.7.

Mayhem 2.7:
- Contains lots of EGTB knowledge. The most important 7-6-5 men.
- I created EGTB generator in Python which I'm not publishing ever. It's for Mayhem only. I will generate more knowledge for v2.8
- Lots of bugs were fixed.
- Speedups
- Now RAM is used optimally.

Release: https://github.com/SamuraiDangyo/mayhem ... s/tag/v2.7
Hi,

are you able to provide an exe or different exes?
gbanksnz at gmail.com
JohnWoe
Posts: 491
Joined: Sat Mar 02, 2013 11:31 pm

Re: Mayhem NNUE - New NN engine

Post by JohnWoe »

I removed Mayhem 2.7 due to under promotion bug. My apologies :D
And released fixed Mayhem 2.71
Release: https://github.com/SamuraiDangyo/mayhem ... /tag/v2.71

Kicked in some speedups as well.
I forgot to mention: Mayhem's evals are now much better. For end users.

Mayhem 2.71 should be around 2800+ Elo.

[pgn][Event "?"]
[Site "?"]
[Date "2021.01.10"]
[Round "2"]
[White "Fruit 2.1"]
[Black "Mayhem 2.71"]
[Result "0-1"]
[TimeControl "60+0.1"]
[ECO "A04"]
[Opening "Reti Opening"]
[PlyCount "158"]
[Annotator "1. +0,10 7... -0,07"]

1. Nf3 {+0,10/12 3}
{s}
1... c5 {+0,00/1}
{s}
2. e4 {+0,42/11 1,8}
{s}
2... d6 {+0,00/1}
{s}
3. Nc3 {+0,72/11 2,0}
{s}
3... Nf6 {+0,00/1}
{s}
4. Bc4 {+0,64/11 2,4}
{s}
4... Nc6 {+0,00/1}
{s}
5. O-O {+0,69/10 1,2}
{s}
5... Bg4 {+0,00/1}
{s}
6. h3 {+0,68/11 1,6}
{s}
6... Bh5 {+0,00/1}
{s}
7. d3 {+0,79/10 1,6}
{s}
7... e6 {-0,07/10 2,1}
{s}
8. Bf4 {+0,56/10 1,5}
{s}
8... a6 {+0,03/9 2,0}
{s}
9. a3 {+0,72/9 2,3}
{s}
9... Nd7 {+0,06/9 2,0}
{s}
10. g4 {+0,37/9 1,9}
{s}
10... Bg6 {+0,03/10 1,9}
{s}
11. Qd2 {+0,21/9 2,7}
{s}
11... Nde5 {+0,24/10 1,9}
{s}
12. Nxe5 {+0,04/10 1,3}
{s}
12... dxe5 {+0,26/10 1,8}
{s}
13. Bg5 {+0,07/10 1,0}
{s}
13... Qc7 {+0,32/10 1,7}
{s}
14. f4 {+0,42/9 1,1}
{s}
14... h6 {+0,48/8 1,7}
{s}
15. f5 {+0,65/10 1,1}
{s}
15... Bh7 {-0,72/9 1,6}
{s}
16. fxe6 {+1,29/10 1,3}
{s}
16... fxe6 {-0,57/10 1,6}
{s}
17. Bh4 {+1,17/10 1,1}
{s}
17... Bg8 {-0,44/8 1,5}
{s}
18. Kh1 {+1,26/8 1,2}
{s}
18... Bd6 {-0,04/9 1,5}
{s}
19. Rf2 {+1,32/8 1,2}
{s}
19... b5 {+0,00/10 1,4}
{s}
20. Ba2 {+1,10/10 1,2}
{s}
20... g5 {-0,08/9 1,4}
{s}
21. Bg3 {+1,02/9 6}
{s}
21... O-O-O {-0,15/9 1,3}
{s}
22. a4 {+1,21/9 9}
{s}
22... b4 {-0,07/10 1,3}
{s}
23. Ne2 {+0,86/9 9}
{s}
23... Na5 {-0,02/9 1,3}
{s}
24. Raf1 {+1,16/8 6}
{s}
24... c4 {-0,04/9 1,2}
{s}
25. Rf3 {+1,01/8 1,3}
{s}
25... Rh7 {-0,01/8 1,2}
{s}
26. Bf2 {+1,15/8 1,2}
{s}
26... Rhd7 {+0,43/9 1,2}
{s}
27. b3 {+1,14/8 7}
{s}
27... c3 {+0,50/10 1,1}
{s}
28. Qc1 {+0,93/10 9}
{s}
28... Be7 {+0,68/10 1,1}
{s}
29. Be3 {+0,94/11 8}
{s}
29... Kb7 {+0,82/11 1,1}
{s}
30. R3f2 {+0,62/10 1,2}
{s}
30... Nc6 {+0,92/10 1,0}
{s}
31. Ng3 {+0,73/11 9}
{s}
31... Nd4 {+1,11/10 10}
{s}
32. Nh5 {+0,70/11 7}
{s}
32... Nxc2 {+1,65/10 10}
{s}
33. Rxc2 {+0,66/9 7}
{s}
33... Rxd3 {+1,48/10 9}
{s}
34. Rcf2 {+0,29/9 9}
{s}
34... Rxe3 {+2,48/9 9}
{s}
35. Qxe3 {+0,58/10 5}
{s}
35... Bc5 {+1,15/10 9}
{s}
36. Qc1 {+0,49/11 7}
{s}
36... Bxf2 {+1,22/10 8}
{s}
37. Rxf2 {+0,28/10 4}
{s}
37... Qd7 {+1,04/10 8}
{s}
38. Ng3 {+0,23/11 8}
{s}
38... Qd1+ {+1,55/10 8}
{s}
39. Qxd1 {+0,21/13 6}
{s}
39... Rxd1+ {+1,04/11 8}
{s}
40. Kg2 {+0,00/14 5}
{s}
40... Kb6 {+1,04/10 7}
{s}
41. Nh5 {+0,00/12 4}
{s}
41... Kc5 {+1,70/10 7}
{s}
42. Ng3 {-0,11/11 5}
{s}
42... Kd4 {+2,04/11 7}
{s}
43. Kh2 {-1,76/10 5}
{s}
43... Kd3 {+3,05/10 7}
{s}
44. Rf3+ {-1,99/11 5}
{s}
44... Kc2 {+2,58/12 7}
{s}
45. Nf1 {-2,37/12 5}
{s}
45... Kb2 {+3,96/12 6}
{s}
46. Rf2+ {-2,86/11 3}
{s}
46... c2 {+4,35/13 6}
{s}
47. Ne3 {-2,88/13 6}
{s}
47... Rc1 {+4,70/13 6}
{s}
48. Kg3 {-3,11/12 5}
{s}
48... Kxa2 {+5,71/13 6}
{s}
49. Rxc2+ {-3,23/13 4}
{s}
49... Rxc2 {+6,51/15 6}
{s}
50. Nxc2 {-4,41/17 4}
{s}
50... Kxb3 {+7,09/14 6}
{s}
51. Ne3 {-4,79/16 3}
{s}
51... Ka2 {+4,28/15 5}
{s}
52. Nc4 {-3,70/14 3}
{s}
52... b3 {+4,47/16 5}
{s}
53. Nd2 {-4,88/15 2,8}
{s}
53... Bf7 {+6,32/10 5}
{s}
54. Kf3 {-5,17/15 5}
{s}
54... a5 {+4,65/14 5}
{s}
55. h4 {-5,66/16 4}
{s}
55... gxh4 {+5,90/16 5}
{s}
56. Kg2 {-6,24/16 3}
{s}
56... b2 {+5,56/15 5}
{s}
57. Kh3 {-9,36/17 5}
{s}
57... Bg6 {+5,86/16 5}
{s}
58. Kxh4 {-6,65/14 4}
{s}
58... b1=Q {+5,83/16 5}
{s}
59. Nxb1 {-10,54/14 4}
{s}
59... Kxb1 {+5,87/16 4}
{s}
60. g5 {-10,76/15 3}
{s}
60... h5 {+6,68/16 4}
{s}
61. Kg3 {-10,19/15 5}
{s}
61... Bxe4 {+6,67/15 4}
{s}
62. Kf2 {-14,33/17 2,8}
{s}
62... Kb2 {+6,79/15 4}
{s}
63. Ke3 {-8,98/11 3}
{s}
63... Bf5 {+6,81/15 4}
{s}
64. Kf3 {-14,32/13 10}
{s}
64... Ka3 {+6,81/14 4}
{s}
65. g6 {-15,46/15 1,8}
{s}
65... Bxg6 {+12,14/14 4}
{s}
66. Kg3 {-15,29/12 2,1}
{s}
66... Kxa4 {+12,22/14 4}
{s}
67. Kh4 {-16,68/13 1,6}
{s}
67... Kb3 {+14,85/13 4}
{s}
68. Kg5 {-21,53/16 1,4}
{s}
68... Bf5 {+19,06/13 3}
{s}
69. Kxh5 {-15,73/10 10}
{s}
69... a4 {+20,02/13 3}
{s}
70. Kh4 {-17,41/11 1,7}
{s}
70... a3 {+20,38/13 3}
{s}
71. Kh5 {-25,10/9 1,0}
{s}
71... a2 {+20,33/12 3}
{s}
72. Kg5 {-25,10/9 1,0}
{s}
72... a1=Q {+28,65/11 3}
{s}
73. Kf6 {-25,30/8 1,0}
{s}
73... e4+ {+28,91/9 3}
{s}
74. Ke7 {-25,50/8 10}
{s}
74... Qg7+ {+30,21/9 3}
{s}
75. Kd6 {-25,60/7 6}
{s}
75... Kc4 {+30,26/9 3}
{s}
76. Kc6 {-25,16/4}
{s}
76... Qa7 {+104,85/7 4}
{s}
77. Kd6 {-25,39/4 0,3}
{s}
77... e3 {+104,85/6 1,8}
{s}
78. Kc6
{-M4/34 0.070s}
78... e5 {+104,85/4 0,5}
{s}
79. Kd6
{-M2/4 0s}
79... Qc5#
{+104.85/3 0s, Black mates} 0-1[/pgn]

Graham Banks wrote: Sat Jan 09, 2021 10:04 pm
JohnWoe wrote: Sat Jan 09, 2021 5:43 pm Mayhem 2.7 released.

Should be much much stronger than any other version.
Now I test without adjudication.
Thanks to Mr. Sylwy's tournament games!
Even badly handicapped Mayhem is 70% / 150 Elo stronger than Fruit 2.1.

Mayhem 2.7 crushes Hakkapeliitta 3.0 on this Linux box even while badly handicapped. (Doesn't have all NNUE flags. Speed is 30% of optimum.)
There is no way Hakkapeliitta is stronger than Mayhem 2.7.

Mayhem 2.7:
- Contains lots of EGTB knowledge. The most important 7-6-5 men.
- I created EGTB generator in Python which I'm not publishing ever. It's for Mayhem only. I will generate more knowledge for v2.8
- Lots of bugs were fixed.
- Speedups
- Now RAM is used optimally.

Release: https://github.com/SamuraiDangyo/mayhem ... s/tag/v2.7
Hi,

are you able to provide an exe or different exes?
Hello!
My compiles on this AMD machine are pretty slow. I might add them later if they perform well.
JohnWoe
Posts: 491
Joined: Sat Mar 02, 2013 11:31 pm

Re: Mayhem NNUE - New NN engine

Post by JohnWoe »

Mayhem 2.8: https://github.com/SamuraiDangyo/mayhem ... s/tag/v2.8

I decided to release this version as it performs really well in my tests.
- More EG code/data. <- Plugging endgames holes.
- Speedups. Some C++ magic here and there.
- Mayhem w/o NNUE is stronger than "FairyMax 5.0b". Plays horribly tho.
- Only Knight promotions which give check are considered during search.
- I guess +20-50 Elo jump.

Mayhem 2.8 vs Fruit. Games played to the bitter end. Mayhem 2.8 can finish games pretty well.

Code: Select all

Score of Mayhem 2.8 vs Fruit 2.1: 30 - 5 - 3  [0.829] 38
Elo difference: 274.16 +/- 157.61
Finished match
JohnWoe
Posts: 491
Joined: Sat Mar 02, 2013 11:31 pm

Re: Mayhem NNUE - New NN engine

Post by JohnWoe »

Mayhem 2.9: https://github.com/SamuraiDangyo/mayhem ... s/tag/v2.9

+ More EG knowledge
+ All simple endgames(KQKR, KNBK ...) tested against SF12. Mayhem converted/held what it should.
+ Fast binaries included.

Should be the best Mayhem.
Sometimes it crushes Glaurung 2.2 in blitz games 90%. Sometimes closer affair.
In Chess960 Mayhem is much weaker. 2600-2700 Elo. Because lacking good 960 Polyglot book?

Nice game proving you need KQKR stuff.
[pgn][Event "Computer Chess Game"]
[Site "pc"]
[Date "2021.01.23"]
[Round "6"]
[White "Hakkapeliitta 3.0"]
[Black "Mayhem 2.9"]
[Result "*"]
[TimeControl "40/60"]

1. e4
{+0,12/19}
1... e5 2. d4
{+0,40/18 3}
2... exd4 3. Nf3
{+0,28/17 2,0}
3... Nc6 4. Nxd4
{+0,37/19 1,9}
4... Qf6 5. Nf3
{+0,51/17 2,8}
5... Bb4+
{-0,37/10 2,0}
6. c3
{+0,25/17 1,9}
6... Bc5
{-0,24/11 1,9}
7. Qd5
{+0,20/17 2,1}
7... d6
{+0,12/10 1,9}
8. Bg5
{-0,19/19 1,9}
8... Qg6
{+0,14/10 1,8}
9. Nh4
{-0,33/20 2,1}
9... Qe6
{+0,13/11 1,7}
10. Qxe6+
{-0,36/19 1,4}
10... Bxe6
{+0,18/12 1,7}
11. Nf5
{-0,56/19 1,3}
11... Bxf5
{+0,33/11 1,6}
12. exf5
{-0,45/16 1,3}
12... f6
{+0,15/10 1,6}
13. Bd2
{-0,63/17 1,2}
13... Nge7
{+0,43/11 1,6}
14. Bd3
{-0,49/17 1,2}
14... O-O-O
{+0,44/11 1,6}
15. b4
{-0,46/17 1,1}
15... Ne5
{+0,43/11 1,6}
16. Bc2
{-0,40/18 1,1}
16... Bb6
{+0,30/12 1,6}
17. a4
{-0,36/17 1,0}
17... a6
{+0,40/11 1,6}
18. O-O
{-0,12/16 1,0}
18... g6
{+0,24/10 1,6}
19. a5
{-0,19/18 1,0}
19... Ba7
{+0,48/12 1,6}
20. fxg6
{-0,29/18 1,0}
20... hxg6
{+0,54/12 1,6}
21. Bf4
{-0,44/19 1,2}
21... g5
{+0,62/11 1,6}
22. Bg3
{-0,93/18 1,0}
22... f5
{+1,36/12 1,6}
23. Bxe5
{-0,97/20 1,0}
23... dxe5
{+1,34/13 1,6}
24. Na3
{-1,27/20 1,2}
24... e4
{+1,60/11 1,6}
25. Nc4
{-1,31/19 1,2}
25... Nd5
{+1,47/11 1,6}
26. Ra3
{-1,39/18 1,0}
26... Nf4
{+1,66/10 1,6}
27. Rd1
{-1,47/20 1,0}
27... Rxd1+
{+1,41/10 1,6}
28. Bxd1
{-1,47/7 0,1}
28... g4
{+1,20/10 1,6}
29. Ne3
{-1,19/19 1,2}
29... Bxe3
{+1,20/12 1,6}
30. fxe3
{-1,14/22 1,0}
30... Nd5
{+0,81/12 1,6}
31. Kf2
{-0,93/21 1,0}
31... Kd7
{+1,18/12 1,6}
32. Bb3
{-1,32/21 1,2}
32... Nxc3
{+0,59/12 1,6}
33. Bc4
{-1,05/21 1,0}
33... Nd1+
{+0,68/12 1,6}
34. Kg3
{-0,93/20 1,0}
34... Ke7
{+0,59/12 1,6}
35. b5
{-0,99/23 2,0}
35... Nb2
{+0,14/12 1,6}
36. Rc3
{-0,77/19 0,9}
36... Kd6
{+0,15/12 1,6}
37. bxa6
{-0,61/22 0,9}
37... Nxc4
{+0,50/14 1,6}
38. Rxc4
{-0,33/23 0,9}
38... bxa6
{+0,59/15 1,6}
39. Rd4+
{-0,92/24 2,0}
39... Kc5
{+0,61/15 1,6}
40. Rd2
{-0,99/23 0,6}
40... Kb4
{+1,53/14 1,6}
41. Rd4+
{-0,97/25 3}
41... Kxa5
{+1,90/16 2,0}
42. Rd5+
{-1,40/25 2,6}
42... Kb4
{+2,72/15 1,9}
43. Rxf5
{-1,79/26 7}
43... c5
{+2,71/14 1,9}
44. Rf7
{-2,10/26 4}
44... c4
{+3,47/14 1,8}
45. Rb7+
{-2,13/24 1,7}
45... Kc3
{+3,82/14 1,7}
46. Kxg4
{-2,42/22 1,8}
46... Kd3
{+3,63/14 1,7}
47. Rd7+
{-2,86/23 2,1}
47... Kxe3
{+3,74/14 1,6}
48. Rc7
{-2,92/24 1,5}
48... Kd3
{+4,09/13 1,6}
49. Rd7+
{-3,37/22 1,6}
49... Kc2
{+2,94/14 1,5}
50. Kf4
{-2,69/21 1,5}
50... c3
{+3,04/12 1,5}
51. Kxe4
{-3,23/21 2,6}
51... Rxh2
{+3,03/12 1,4}
52. Kf3
{-3,49/20 1,2}
52... Rh6
{+3,15/12 1,4}
53. Rb7
{-4,17/21 1,2}
53... Rc6
{+3,97/11 1,4}
54. g4
{-3,90/24 1,1}
54... a5
{+3,45/12 1,4}
55. g5
{-4,36/20 1,1}
55... Ra6
{+3,17/13 1,4}
56. Rf7
{-4,34/23 1,4}
56... Rg6
{+3,09/12 1,4}
57. Kg4
{-3,65/21 1,0}
57... a4
{+2,77/12 1,4}
58. Kf5
{-2,58/21 1,1}
58... Rg8
{+2,60/14 1,4}
59. Ra7
{-3,61/21 0,9}
59... Kb3
{+2,80/14 1,4}
60. Rb7+
{-3,51/25 0,9}
60... Ka2
{+2,60/14 1,4}
61. Rc7
{-3,44/25 0,9}
61... Kb2
{+2,12/14 1,4}
62. Rb7+
{-3,22/24 0,9}
62... Ka3
{+2,06/15 1,4}
63. Rc7
{-3,20/25 1,1}
63... Rf8+
{+1,85/14 1,4}
64. Ke6
{-2,92/22 0,9}
64... Kb2
{+1,91/13 1,4}
65. Rb7+
{-3,00/24 1,1}
65... Ka2
{+1,35/13 1,4}
66. Rc7
{-2,03/23 0,9}
66... Rf3
{+1,26/13 1,4}
67. g6
{-1,11/23 0,9}
67... Rg3
{+1,40/13 1,4}
68. Kf7
{+0,00/25 1,0}
68... Kb2
{+1,25/14 1,4}
69. Rb7+
{-0,21/25 1,9}
69... Ka3
{+1,05/14 1,4}
70. Rc7
{-0,21/27 1,2}
70... Rf3+
{+1,09/13 1,4}
71. Ke7
{-0,21/25 0,9}
71... Kb2
{+1,20/13 1,4}
72. Rb7+
{-0,21/25 0,9}
72... Ka2
{+1,02/14 1,4}
73. Rc7
{-0,21/25 0,9}
73... Kb3
{+0,83/13 1,4}
74. Rb7+
{-0,27/25 1,6}
74... Kc2
{+0,84/14 1,4}
75. g7
{+0,00/26 0,8}
75... Re3+
{+0,66/13 1,4}
76. Kf6
{+0,00/26 0,8}
76... Rg3
{+0,62/14 1,4}
77. Kf7
{+0,00/25 0,8}
77... Kd3
{+0,54/13 1,4}
78. Rd7+
{+0,00/25 0,8}
78... Kc4
{+0,67/13 1,4}
79. g8=Q
{+0,00/28 0,9}
79... Rxg8
{+0,96/15 1,4}
80. Kxg8
{+0,00/29 0,8}
80... c2
{+1,11/14 1,4}
81. Rc7+
{-3,96/33 2,5}
81... Kb3
{+0,81/15 2,0}
82. Rb7+
{-4,25/34 2,3}
82... Ka2
{+0,95/15 1,9}
83. Rc7
{-4,32/36 2,2}
83... Kb1
{+0,98/15 1,9}
84. Rb7+
{-4,32/36 2,1}
84... Ka1
{+0,67/16 1,8}
85. Rc7
{-4,32/37 2,1}
85... Kb2
{+0,93/16 1,7}
86. Rb7+
{-4,32/37 2,0}
86... Kc1
{+0,81/15 1,7}
87. Ra7
{-4,33/36 1,9}
87... Kd1
{+0,69/14 1,6}
88. Rd7+
{+0,00/42 1,8}
88... Ke2
{+0,00/17 1,6}
89. Rc7
{+0,00/47 1,9}
89... Kd3
{+0,00/18 1,5}
90. Rd7+
{+0,00/45 1,7}
90... Ke3
{+0,00/18 1,5}
91. Re7+
{+0,00/49 1,6}
91... Kd2
{+0,00/18 1,4}
92. Rd7+
{+0,00/54 1,6}
92... Kc3
{+0,00/18 1,4}
93. Rc7+
{+0,00/57 1,6}
93... Kd2
{+0,00/18 1,4}
94. Rd7+
{+0,00/79 7}
94... Ke1
{+0,00/18 1,4}
95. Rc7
{+0,00/58 1,2}
95... Kd1
{+0,00/18 1,4}
96. Rd7+
{+0,00/86 5}
96... Kc1
{+0,00/19 1,4}
97. Ra7
{+0,00/55 4}
97... a3
{+0,57/15 1,4}
98. Kf8
{-4,33/27 0,7}
98... Kb2
{+1,37/14 1,4}
99. Rb7+
{-4,33/30 0,7}
99... Kc3
{+1,64/14 1,4}
100. Rc7+
{-4,33/30 0,7}
100... Kb3
{+1,00/15 1,4}
101. Rb7+
{-4,33/30 0,7}
101... Kc4
{+0,83/15 1,4}
102. Rc7+
{-4,34/27 1,0}
102... Kd3
{+0,92/15 1,4}
103. Rd7+
{-4,33/28 0,7}
103... Ke3
{+1,13/14 1,4}
104. Re7+
{-4,33/27 0,7}
104... Kd4
{+0,84/15 1,4}
105. Rc7
{-4,33/26 0,7}
105... a2
{+0,99/14 1,4}
106. Rxc2
{-4,34/26 0,7}
106... a1=Q
{+1,02/12 1,4}
107. Rd2+
{-4,34/25 0,7}
107... Ke5
{+10,15/13 1,4}
108. Re2+
{-4,34/25 0,7}
108... Kf5
{+10,15/13 1,4}
109. Rf2+
{-4,44/25 0,7}
109... Kg6
{+10,15/13 1,4}
110. Rg2+
{-4,34/28 0,7}
110... Kh7
{+10,44/13 1,4}
111. Rh2+
{+0,00/62 0,7}
111... Kg6 112. Rg2+
{+0,00/106 4}
112... Kh6
{+9,91/13 1,6}
113. Rh2+
{-4,34/24 0,4}
113... Kg5
{+9,92/13 1,6}
114. Rg2+
{-4,39/24 0,4}
114... Kh4
{+9,98/13 1,6}
115. Rg6
{-4,36/22 0,5}
115... Qe5
{+10,35/13 1,6}
116. Kf7
{-4,39/22 0,5}
116... Qf5+
{+10,35/13 1,6}
117. Rf6
{-4,44/24 0,5}
117... Qd7+
{+10,36/13 1,6}
118. Kg6
{-4,44/26 0,6}
118... Kg4
{+10,37/13 1,6}
119. Rf7
{-4,47/27 0,5}
119... Qd3+
{+10,29/12 1,6}
120. Kg7
{-4,47/28 0,5}
120... Kg5
{+10,22/11 1,6}
121. Re7
{-8,98/31 2,4}
121... Qd5
{+10,15/12 2,0}
122. Rf7
{-1000,14/32 6}
122... Qc5
{+10,12/12 1,9}
123. Rd7
{-1000,19/33 2,1}
123... Qc6
{+9,90/12 1,9}
124. Rf7
{-1000,18/34 2,0}
124... Qe8
{+9,82/12 1,8}
125. Rf1
{-1000,17/35 1,9}
125... Qe7+
{+9,60/11 1,8}
126. Kg8
{-1000,16/32 1,8}
126... Kh5
{+9,58/11 1,7}
127. Rh1+
{-4,35/29 1,8}
127... Kg4
{+9,29/14 1,6}
128. Rg1+
{-4,46/28 1,7}
128... Kf5
{+9,08/13 1,6}
129. Rf1+
{-4,47/29 1,6}
129... Kg6
{+8,28/17 1,5}
130. Rg1+
{-4,98/35 2,1}
130... Kh6
{+8,37/16 1,5}
131. Rh1+
{+0,00/76 1,5}
131... Kg5
{+8,16/15 1,4}
132. Rf1
{+0,00/101 7}
132... Qb4
{+8,71/11 1,4}
133. Rg1+
{-8,32/26 1,1}
133... Kf6
{+8,49/13 1,4}
134. Rf1+
{-9,16/28 1,1}
134... Ke7
{+8,52/13 1,4}
135. Rf7+
{-9,48/28 1,1}
135... Ke8
{+8,39/13 1,4}
136. Rg7
{-11,18/29 2,5}
136... Qh4
{+8,63/11 1,4}
137. Rg6
{-9,73/24 1,0}
137... Qe7
{+8,29/12 1,4}
138. Kh8
{-9,25/27 0,9}
138... Qe5+
{+7,91/12 1,4}
139. Kg8
{-1000,13/28 0,9}
139... Ke7
{+7,95/12 1,4}
140. Rg1
{-1000,12/30 0,9}
140... Qb8+
{+8,04/11 1,4}
141. Kg7
{-1000,17/30 0,9}
141... Qf8+
{+7,70/12 1,4}
142. Kh7
{-1000,14/30 0,9}
142... Qf5+
{+7,55/12 1,4}
143. Kg7
{-1000,13/29 0,9}
143... Qf7+
{+7,36/12 1,4}
144. Kh6
{-1000,14/25 0,9}
144... Qf4+
{+14,53/13 1,4}
145. Kg6
{-1000,14/25 0,9}
145... Qd6+
{+14,59/12 1,4}
146. Kf5
{-1000,10/24 0,9}
146... Qc5+
{+14,54/12 1,4}
147. Ke4
{-1000,09/23 0,9}
147... Qxg1
{+14,90/12 1,4}

*

[/pgn]
User avatar
Sylwy
Posts: 4466
Joined: Fri Apr 21, 2006 4:19 pm
Location: IASI - the historical capital of MOLDOVA
Full name: SilvianR

Re: Mayhem NNUE - New NN engine

Post by Sylwy »

It will be the moment of truth !

Image
JohnWoe
Posts: 491
Joined: Sat Mar 02, 2013 11:31 pm

Re: Mayhem NNUE - New NN engine

Post by JohnWoe »

Sylwy wrote: Sat Jan 23, 2021 9:37 pm It will be the moment of truth !

Image
Thanks!

Go Mayhem! :D
Will be interesting to see results.
My tests Mayhem 2.9 vs Glaurung 2.2 both used 256MB but Mayhem used its own book. Good results but maybe book is worth so much Elo. I don't know. I just feel opening book is essential for any game playing engine. Mayhem has now tons of (hand picked) endgames built into too.