The Mystic And The Cosmos

dkappe · Post by **dkappe** » Sun May 23, 2021 7:30 pm

Graham Banks wrote: ↑Fri May 21, 2021 2:34 pm I'm using the SSE binary.
I think that engines should be released with non-popcount, popcount, and bmi2 exes at least.

Toga II predates most of these instruction sets. Only the nnue library takes advantage of them. So toga III needs to be reworked substantially to take advantage of them.

I’ve started a gauntlet for Halogen 10 vs Toga III 0.2 and 0.3.12, all with avx2. So far they bracket Halogen. We’ll see where it ends up after a few thousand games.

Graham Banks · Post by **Graham Banks** » Tue May 25, 2021 12:04 am

THE MYSTIC AND THE COSMOS

Intel i5 Quad
ChessGUI
256mb hash each
3-4-5 piece tablebases
Ponder off
GBSelect2019.cgb book
40 moves in 25 minutes repeating (adapted for the CCRL)
2 cycles 26 rounds
All engines 64-bit where available

Standings after Round 16

11.0 - Halogen 10 64-bit
11.0 - Schooner 2.2 64-bit
11.0 - Marvin 5.0.0 64-bit
9.5 - Winter 0.9 64-bit
9.0 - Clover 2.3.1 64-bit
9.0 - Lc0 0.27.0 w703810 64-bit
8.5 - Arasan 22.2 64-bit
8.0 - Roc 1.7 64-bit
6.5 - Seer 2.0.1 64-bit
6.5 - Wasp 4.50 64-bit
6.5 - Combusken 1.4.0 64-bit
6.0 - Orion 0.8 64-bit
6.0 - Nirvanachess 2.5 64-bit
3.5 - Toga III 0.3.12 64-bit

Games and standings will be available after every four rounds from the following link:
http://kirill-kryukov.com/chess/discuss ... =7&t=13008

If you install TLCV (Tom's Live Chess Viewer) on your computer, you can watch the games live move by move. You'll also be able to chat to others following the tournament in the chatroom there.
http://kirill-kryukov.com/chess/discuss ... p?id=42959
Host - GrahamCCRL.dyndns.org Port - 16053

Linux users can use Livius:
https://github.com/kmar/livius

There is also a Livius windows version.
It has live pv boards as a nice addition.
http://www.crabaware.com/livius/

dkappe · Post by **dkappe** » Tue May 25, 2021 12:09 am

Very odd. With avx2 at 60+1 I get something different. I have to question the configuration.

Code: Select all


   # PLAYER        :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)     W     D    L  D(%)
   1 Halogen10     :       0   ----  2007.5    3611  55.6     100  1367  1281  963  35.5
   2 toga3.3.12    :     -22     12   847.5    1806  46.9     100   525   645  636  35.7
   3 toga3.2       :     -58     13   756.0    1805  41.9     ---   438   636  731  35.2

White advantage = 44.97 +/- 4.69
Draw rate (equal opponents) = 36.37 % +/- 0.82

connor_mcmonigle · Post by **connor_mcmonigle** » Tue May 25, 2021 12:32 am

dkappe wrote: ↑Tue May 25, 2021 12:09 am Very odd. With avx2 at 60+1 I get something different. I have to question the configuration.

Code: Select all


   # PLAYER        :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)     W     D    L  D(%)
   1 Halogen10     :       0   ----  2007.5    3611  55.6     100  1367  1281  963  35.5
   2 toga3.3.12    :     -22     12   847.5    1806  46.9     100   525   645  636  35.7
   3 toga3.2       :     -58     13   756.0    1805  41.9     ---   438   636  731  35.2

White advantage = 44.97 +/- 4.69
Draw rate (equal opponents) = 36.37 % +/- 0.82

This discrepancy is easily explained by Graham's CPU lacking AVX2, no? It's a first gen, slow i5 so an underperformance from Toga-NNUE is fairly expected and the sample size remains quite small as well. Given these factors, I think it's a bit premature to suggest the configuration is somehow flawed.

In any case, I've just started a match between Halogen and your latest Toga-NNUE (both natively compiled on Linux; I'm using the AVX2 compile) at 1m+1s on 1 thread with 128mb hash and no EGTB for both. Toga-NNUE reports about 1.4Mnps and Halogen reports about 3.6Mnps from the start position (R5 3600X). I'm using a more standard, somewhat drawish, book which has provided reasonable estimates of relative CCRL elo for me in the past (https://github.com/AndyGrant/OpenBench/ ... v3.pgn.zip). So far Halogen seems quite dominant, but I'll report back when the sample size is larger.

connor_mcmonigle · Post by **connor_mcmonigle** » Tue May 25, 2021 4:10 am

Code: Select all

Score of halogen vs toga-nnue: 115 - 47 - 81 [0.640]
...      halogen playing White: 71 - 13 - 37  [0.740] 121
...      halogen playing Black: 44 - 34 - 44  [0.541] 122
...      White vs Black: 105 - 57 - 81  [0.599] 243
Elo difference: 99.9 +/- 36.5, LOS: 100.0 %, DrawRatio: 33.3 %
248 of 200000 games finished.

I've decided to stop the test here as the above is clearly inconsistent with your results. The book might be relevant, but perhaps I've configured something incorrectly as the above result is worse than CCRL's estimate for "Toga III-0.2" and 0.3 is expected to be stronger than 0.2.

dkappe · Post by **dkappe** » Tue May 25, 2021 9:12 am

connor_mcmonigle wrote: ↑Tue May 25, 2021 4:10 am
Code: Select all
Score of halogen vs toga-nnue: 115 - 47 - 81 [0.640]
...      halogen playing White: 71 - 13 - 37  [0.740] 121
...      halogen playing Black: 44 - 34 - 44  [0.541] 122
...      White vs Black: 105 - 57 - 81  [0.599] 243
Elo difference: 99.9 +/- 36.5, LOS: 100.0 %, DrawRatio: 33.3 %
248 of 200000 games finished.
I've decided to stop the test here as the above is clearly inconsistent with your results. The book might be relevant, but perhaps I've configured something incorrectly as the above result is worse than CCRL's estimate for "Toga III-0.2" and 0.3 is expected to be stronger than 0.2.

If you’re running on Windows that may be the case. I have no way to test that. Toga II (and III) is not up to date as far as recent processors go. I’m just really perplexed that 0.2 seems to outperform 0.3.12 on some platforms.

Graham Banks · Post by **Graham Banks** » Tue May 25, 2021 9:21 am

[pgn][Event "The Mystic And The Cosmos"]
[Site "ChessGUI2"]
[Date "2021.05.25"]
[Round "19.3"]
[White "Toga III 0.3.12 64-bit"]
[Black "Winter 0.9 64-bit"]
[Result "1-0"]
[Time "5:22:35 p.m."]
[ECO "A05"]
[Opening "King's Indian Attack; Reti; Zukertort; A05"]
[TimeControl "40/1500:40/1500:40/1500"]
[PlyCount "125"]
[Number "129"]
[Termination "GUI adjudication"]
[WhiteType "program"]
[BlackType "program"]
[Variant "normal"]

{ Intel i5 Quad }
1.Nf3 {[%eval 0,1] [%emt 00:00:00]} Nf6 {[%eval 0,1] [%emt 00:00:00]}
2.c4 {[%eval 0,1] [%emt 00:00:00]} b6 {[%eval 0,1] [%emt 00:00:00]}
3.Nc3 {[%eval 0,1] [%emt 00:00:00]} Bb7 {[%eval 0,1] [%emt 00:00:00]}
4.d4 {[%eval 0,1] [%emt 00:00:00]} e6 {[%eval 0,1] [%emt 00:00:00]}
5.a3 {[%eval 0,1] [%emt 00:00:00]} d5 {[%eval 0,1] [%emt 00:00:00]}
6.cxd5 {[%eval 0,1] [%emt 00:00:00]} Nxd5 {[%eval 0,1] [%emt 00:00:00]}
7.Qc2 {[%eval 0,1] [%emt 00:00:00]} Nxc3 {[%eval 0,1] [%emt 00:00:00]}
8.Qxc3 {[%eval 0,1] [%emt 00:00:00]} h6 {[%eval 0,1] [%emt 00:00:00]}
9.g3 {[%eval -4,19] [%emt 00:00:30]} Nd7 {(Nd7) [%eval -7,23] [%emt 00:00:34]}
10.Bg2 {(Bg2) [%eval 2,20] [%emt 00:00:32]} Bd6 {(Nf6) [%eval 0,22] [%emt 00:00:28]}
11.d5 {(O-O) [%eval -3,20] [%emt 00:00:46]} Bxd5 {(O-O) [%eval 22,24] [%emt 00:00:32]}
12.Qxg7 {(Qxg7) [%eval 14,22] [%emt 00:00:46]} Qf6 {(Qf6) [%eval 22,26] [%emt 00:00:28]}
13.Qxf6 {(Qxf6) [%eval 12,22] [%emt 00:00:27]} Nxf6 {(Nxf6) [%eval 22,25] [%emt 00:00:57]}
14.O-O {(b4) [%eval 13,21] [%emt 00:00:46]} c5 {(c5) [%eval 3,24] [%emt 00:00:56]}
15.Nd2 {(Nd2) [%eval 17,20] [%emt 00:00:39]} Rd8 {(O-O-O) [%eval 0,23] [%emt 00:00:56]}
16.e4 {(e4) [%eval 15,20] [%emt 00:00:32]} Ba8 {(Ba8) [%eval 5,25] [%emt 00:00:35]}
17.Re1 {(f4) [%eval 23,19] [%emt 00:00:31]} Be5 {(Be5) [%eval -12,22] [%emt 00:00:56]}
18.Nc4 {(Nc4) [%eval 30,22] [%emt 00:00:40]} Bd4 {(Bd4) [%eval -16,25] [%emt 00:00:53]}
19.e5 {(e5) [%eval 16,21] [%emt 00:00:41]} Bxg2 {(Bxg2) [%eval -32,22] [%emt 00:00:31]}
20.Kxg2 {(Kxg2) [%eval 16,21] [%emt 00:00:35]} Nd5 {(Nd5) [%eval -25,25] [%emt 00:00:56]}
21.Bd2 {(Nd6) [%eval 20,21] [%emt 00:00:36]} a6 {(Kd7) [%eval 0,21] [%emt 00:00:32]}
22.Rac1 {(Rac1) [%eval 25,19] [%emt 00:00:36]} b5 {(Ke7) [%eval 0,24] [%emt 00:00:34]}
23.Nd6 {(Nd6) [%eval 19,21] [%emt 00:00:31]} Ke7 {(Ke7) [%eval 0,25] [%emt 00:00:32]}
24.Rc2 {(b3) [%eval 10,20] [%emt 00:00:52]} Rb8 {(f5) [%eval 3,22] [%emt 00:00:30]}
25.Bc1 {(Bc3) [%eval 11,21] [%emt 00:00:48]} a5 {(c4) [%eval 34,21] [%emt 00:01:00]}
26.Ne4 {(Ne4) [%eval 0,20] [%emt 00:00:33]} a4 {(Rbc8) [%eval 29,22] [%emt 00:00:38]}
27.Ree2 {(Rce2) [%eval 0,22] [%emt 00:01:00]} c4 {(Rbc8) [%eval 40,27] [%emt 00:01:01]}
28.f4 {(Nc3) [%eval 5,20] [%emt 00:00:34]} Ba7 {(Rhc8) [%eval 14,23] [%emt 00:00:39]}
29.h4 {(Bd2) [%eval 6,21] [%emt 00:01:26]} h5 {(Kd7) [%eval 16,23] [%emt 00:00:38]}
30.Kh3 {(Bd2) [%eval 15,19] [%emt 00:00:33]} Bg1 {(Bd4) [%eval 5,22] [%emt 00:01:03]}
31.Re1 {(Bd2) [%eval 10,20] [%emt 00:00:56]} Bb6 {(Bd4) [%eval 0,26] [%emt 00:00:38]}
32.Rce2 {(Rf1) [%eval 37,20] [%emt 00:00:38]} Kd7 {(Bc7) [%eval 0,22] [%emt 00:01:04]}
33.f5 {(f5) [%eval 41,22] [%emt 00:00:47]} Rbf8 {(Rbf8) [%eval -43,22] [%emt 00:00:48]}
34.Rf1 {(Rd1) [%eval 38,22] [%emt 00:00:36]} Kc6 {(Kc6) [%eval -46,21] [%emt 00:00:36]}
35.Ng5 {(Ng5) [%eval 56,23] [%emt 00:00:46]} Rhg8 {(Bd8) [%eval -60,23] [%emt 00:01:00]}
36.Bd2 {(Bd2) [%eval 55,22] [%emt 00:01:01]} Nc7 {(Rh8) [%eval -30,24] [%emt 00:00:50]}
37.Be3 {(Kg2) [%eval 124,22] [%emt 00:01:03]} Nd5 {(Nd5) [%eval -64,24] [%emt 00:01:09]}
38.Bxb6 {(Bxb6) [%eval 117,23] [%emt 00:01:07]} Kxb6 {(Kxb6) [%eval -98,25] [%emt 00:00:37]}
39.Rf3 {(fxe6) [%eval 134,23] [%emt 00:01:24]} exf5 {(Rc8) [%eval -85,24] [%emt 00:01:17]}
40.Rxf5 {(Rxf5) [%eval 157,20] [%emt 00:01:05]} c3 {(c3) [%eval -141,22] [%emt 00:00:52]}
41.bxc3 {(bxc3) [%eval 180,21] [%emt 00:00:46]} Rc8 {(Rc8) [%eval -153,23] [%emt 00:00:43]}
42.Rd2 {(Re1) [%eval 184,20] [%emt 00:00:47]} Nxc3 {(Nxc3) [%eval -120,23] [%emt 00:00:34]}
43.Rf6 {(Rf6) [%eval 188,20] [%emt 00:00:45]} Ka5 {(Ka5) [%eval -89,24] [%emt 00:00:31]}
44.Rdd6 {(Rd7) [%eval 158,20] [%emt 00:01:06]} Ra8 {(Ra8) [%eval -101,24] [%emt 00:00:24]}
45.Rf5 {(Rf5) [%eval 138,19] [%emt 00:00:54]} Nb1 {(Nb1) [%eval -124,25] [%emt 00:00:36]}
46.Rf3 {(Rd3) [%eval 132,20] [%emt 00:00:32]} Rgb8 {(Rac8) [%eval -76,24] [%emt 00:00:29]}
47.Nxf7 {(Nxf7) [%eval 155,19] [%emt 00:00:41]} Nxa3 {(Nxa3) [%eval -99,27] [%emt 00:00:38]}
48.Rxa3 {(Rxa3) [%eval 157,19] [%emt 00:00:28]} b4 {(b4) [%eval -106,27] [%emt 00:00:22]}
49.Rf3 {(Re3) [%eval 155,18] [%emt 00:00:36]} b3 {(b3) [%eval -63,24] [%emt 00:00:29]}
50.Rd1 {(Rd1) [%eval 151,19] [%emt 00:00:35]} a3 {(a3) [%eval -71,25] [%emt 00:00:25]}
51.e6 {(e6) [%eval 152,17] [%emt 00:00:34]} a2 {(a2) [%eval -60,25] [%emt 00:00:24]}
52.e7 {(Ne5) [%eval 160,16] [%emt 00:00:25]} Kb4 {(b2) [%eval -130,22] [%emt 00:00:31]}
53.Ne5 {(Ne5) [%eval 227,18] [%emt 00:00:46]} Rc8 {(Rc8) [%eval -129,22] [%emt 00:00:26]}
54.Nd3 {(Nd3) [%eval 280,18] [%emt 00:00:29]} Ka5 {(Ka3) [%eval -159,23] [%emt 00:00:42]}
55.Nb2 {(e8=Q) [%eval 326,18] [%emt 00:00:30]} Kb4 {(Kb4) [%eval -220,22] [%emt 00:00:47]}
56.Rf4 {(Rf4) [%eval 326,18] [%emt 00:00:22]} Kb5 {(Kb5) [%eval -346,23] [%emt 00:00:32]}
57.Nc4 {(Nc4) [%eval 334,18] [%emt 00:00:41]} b2 {(Rg8) [%eval -336,25] [%emt 00:00:28]}
58.Nd6 {(Nd6) [%eval 462,18] [%emt 00:00:35]} Ka5 {(Ka5) [%eval -339,26] [%emt 00:00:46]}
59.Rd5 {(Rd5) [%eval 351,18] [%emt 00:00:28]} Rc5 {(Kb6) [%eval -659,26] [%emt 00:00:28]}
60.Nc4 {(Rxc5) [%eval 559,17] [%emt 00:00:26]} Kb4 {(Kb5) [%eval -944,24] [%emt 00:00:48]}
61.Nxb2 {(Nxb2) [%eval 677,18] [%emt 00:00:23]} Kb3 {(Kb3) [%eval -1150,28] [%emt 00:00:42]}
62.Rxc5 {(Rxc5) [%eval 762,19] [%emt 00:00:11]} a1=Q {(a1=Q) [%eval -1150,30] [%emt 00:00:24]}
63.Rb5 {(Rb5) [%eval 819,20] [%emt 00:00:09]}
1-0[/pgn]

Graham Banks · Post by **Graham Banks** » Thu May 27, 2021 9:04 am

THE MYSTIC AND THE COSMOS

Intel i5 Quad
ChessGUI
256mb hash each
3-4-5 piece tablebases
Ponder off
GBSelect2019.cgb book
40 moves in 25 minutes repeating (adapted for the CCRL)
2 cycles 26 rounds
All engines 64-bit where available

Standings after Round 20

13.5 - Halogen 10 64-bit
13.0 - Schooner 2.2 64-bit
12.0 - Marvin 5.0.0 64-bit
11.0 - Winter 0.9 64-bit
11.0 - Clover 2.3.1 64-bit
11.0 - Lc0 0.27.0 w703810 64-bit
10.0 - Roc 1.7 64-bit
10.0 - Arasan 22.2 64-bit
9.0 - Orion 0.8 64-bit
9.0 - Combusken 1.4.0 64-bit
8.5 - Wasp 4.50 64-bit
8.5 - Nirvanachess 2.5 64-bit
8.0 - Seer 2.0.1 64-bit
5.5 - Toga III 0.3.12 64-bit

Games and standings will be available after every four rounds from the following link:
http://kirill-kryukov.com/chess/discuss ... =7&t=13008

If you install TLCV (Tom's Live Chess Viewer) on your computer, you can watch the games live move by move. You'll also be able to chat to others following the tournament in the chatroom there.
http://kirill-kryukov.com/chess/discuss ... p?id=42959
Host - GrahamCCRL.dyndns.org Port - 16053

Linux users can use Livius:
https://github.com/kmar/livius

There is also a Livius windows version.
It has live pv boards as a nice addition.
http://www.crabaware.com/livius/

Graham Banks · Post by **Graham Banks** » Fri May 28, 2021 1:06 pm

THE MYSTIC AND THE COSMOS

Intel i5 Quad
ChessGUI
256mb hash each
3-4-5 piece tablebases
Ponder off
GBSelect2019.cgb book
40 moves in 25 minutes repeating (adapted for the CCRL)
2 cycles 26 rounds
All engines 64-bit where available

Final Standings

16.5 - Halogen 10 64-bit
14.5 - Winter 0.9 64-bit
14.5 - Schooner 2.2 64-bit
14.0 - Clover 2.3.1 64-bit
14.0 - Marvin 5.0.0 64-bit
14.0 - Combusken 1.4.0 64-bit
13.5 - Roc 1.7 64-bit
13.5 - Lc0 0.27.0 w703810 64-bit
13.5 - Arasan 22.2 64-bit
12.0 - Orion 0.8 64-bit
12.0 - Wasp 4.50 64-bit
11.5 - Nirvanachess 2.5 64-bit
10.0 - Seer 2.0.1 64-bit
8.5 - Toga III 0.3.12 64-bit

The complete tournament pgn (zipped) can be downloaded here:
http://kirill-kryukov.com/chess/discuss ... p?id=48423

Graham Banks · Post by **Graham Banks** » Fri May 28, 2021 1:07 pm

Congratulations to Kieran (Halogen author) for winning a tough tournament.

Thanks to all who took an interest.

The Mystic And The Cosmos

Re: The Mystic And The Cosmos

Re: The Mystic And The Cosmos

Re: The Mystic And The Cosmos

Re: The Mystic And The Cosmos

Re: The Mystic And The Cosmos

Re: The Mystic And The Cosmos

Re: The Mystic And The Cosmos

Re: The Mystic And The Cosmos

Re: The Mystic And The Cosmos

Re: The Mystic And The Cosmos