CEGT - rating lists August 04th 2024

[Werner]

Hi,
our actual rating lists are online and can be found under the attached links!

40 / 20:
New games: 3.100 with 29 different engines
Total:         1.842.966      games

NEW Engines
1 Torch 3 x64 1CPU 3639 +12 -12 2200 games (+37 to v. 2)
22 Caissa 1.20 x64 1CPU 3559 +17 -17 1000 games (+14 to v. 1.19)

UPDATES
50 Lizard 10.5 x64 1CPU 3523 +13 -13 2100 games (+1)
593 Princhess 0.18.0 x64 1CPU 3078 +19 -19 1000 games (- 1)
5 Torch 2 x64 1CPU 3602 +11 -11 3229 games (+3 )

40 / 4
last update was July 26th 2024
New games: 29.100; total now: 3.905.430 games (3242 engines, +12)

we are testing:
Halogen 12.0NN x64 1CPU
Caissa 1.20NN x64 1CPU: ~ELO 3599 / 1200 games => +23 to v. 1.18 (3576)
Torch 3NN x64 1CPU: ~ELO 3661 / 2500 games

Lizard 10.5NN x64 1CPU: ~ ELO 3553 / 2000 games => +59 to v. 10.3 (3494)
Odonata 1.0.0NN x64 1CPU: ~ ELO 3181 / 2300 games => +150 to v. 0.9.0 (3031)
Princhess 0.18.0 x64 1CPU: ~ ELO 3107 / 1600 games => +56 to v. 0.17.0 (3051)
Patricia 3.0NN x64 1CPU: ~ ELO 3181 / 1700 games => +200 to v. 2.0.1NN (2981)
Nalwald 19 x64 1CPU: ~ ELO 3216 / 2000 games => +57 to v. 18 (3159)
Sirius 7.0 x64 1CPU: ~ ELO 3121 / 2600 games -new-
Renegade 1.1.0NN x64: ~3446 / 1600 / +120 (1.0.0 = 3326)
Viridithas 13.0.0NN x64 1CPU: ~3531 / 2000 / +41 to v12.0.0
Obsidian 13.0NN x64 1CPU: ~ ELO 3616 / 2700 games => +20 to v. 12.0 (3596)
Winter 4.0NN x64 1CPU: ~ ELO 3372 / 2200 games => +140 to v. 3.0 (3232)
KnightX 4.0 x64: ~ 2587 / 1500 games / +29 to v3.8 (2558, v3.9 not tested)
Stash 36.0 x64 1CPU: ~ ELO 3257 / 2100 games => +34 to v. 35.0 (3223)
Stormphrax 5.0.0NN x64 1CPU: ~ ELO 3499 / 2400 games => +59 to v. 4.0.0 (3440, 4.1.0 not tested)
Lynx 1.5.1 x64: ~2660 / 1500 / NEW
Pedantic 1.1.0 x64 1CPU: ~ ELO 3046 / 1900 games => +31 to v. 1.0.0 (3015)
Tucano 11.00.1NN x64 1CPU: ~ 3264 / 2000 games => -14 to v10.00

5'+3'' pb=on
last update is from July 10th with 6.000 new games, total now 534.450 with 399 Engines/Versions (+3)

we are testing:
Testing Torch 3NN: ~3653 /2200 games

Stash 36.0 x64: ~3270 / 1500 / +13 to 35.0 (3257)
Wasp 7.00NN x64: ~3428 /1100
Texel 1.11NN x64: ~ 3407 / 1400 games => currently NOT enough for the MainList (should be ~~3440)
LC0 0.31 dist-swa 3395000: ~ 3597/900
Dumb 2.1 x64: (later)
and: https://cegt.forumieren.com/t2229-tourn ... -list-only

A big „Thank you“ to all testers as usual!!

Links

40/20: http://www.cegt.net/rating.htm
Blitz: http://www.cegt.net/blitz.htm
40/120: http://www.cegt.net/rating120.htm
25+8: http://www.cegt.net/rating25plus8.htm
3+1 pb=on: http://www.cegt.net/rating3plus1pbon.htm
5+3 pb=on: http://www.cegt.net/rating5plus3pbon.htm
Tester: http://www.cegt.net/testers/testers.htm
Games of the week: http://www.cegt.net/40_40%20Rating%20Li ... on/gow.jpg

Werner
CEGT-Team

[Jouni]

1 Torch 3 x64 1CPU 3639 12 12 2200 64.1% 3533 63.8%
2 Stockfish 16.1 x64 1CPU 3623 10 10 3566 62.6% 3529 73.4%

But Torch loses badly in match vs Stockfish .

[Werner]

1 Torch 3 x64 1CPU 3639 12 12 2200 64.1% 3533 63.8%
2 Stockfish 16.1 x64 1CPU 3623 10 10 3566 62.6% 3529 73.4%
But Torch loses badly in match vs Stockfish .

We have no easy answer: as on 40/4 list we have
2 Stockfish 16.1NN x64 1CPU 3654 10 10 3500
and Torch3 about ELO 3661 / 2500 games (49,5% against SF 16.1).
We use same opening set, 40/4 random and 40/20 a fixed set.
I use for 40/20 the AVX512 compile, rest of the team the AVX2 compile.
Perhaps my set was good for Torch? I changed it now to openings which a Chess Player would use too.

[Jouni]

Simply Torch plays better against weaker engines meaning better rating?

[Werner]

I think the used openings causes the differences, e.g.:
with unbalanced openings
1 Torch 3 x64 1CPU +135 +40/=57/-3 68.50% 68.5/100
2 Caissa 1.19 x64 1CPU -135 +3/=57/-40 31.50% 31.5/100

1 Stockfish 16.1 x64 1CPU +53 +21/=73/-6 57.50% 57.5/100
2 Torch 3 x64 1CPU -53 +6/=73/-21 42.50% 42.5/100

with more balanced openings (about 0,3 and +-0.3)
1 Torch 3 x64 1CPU +70 +20/=80/-0 60.00% 60.0/100
2 Caissa 1.19 x64 1CPU -70 +0/=80/-20 40.00% 40.0/100

with totaly balanced openings (0.00 )
1 Stockfish 16.1 x64 1CPU +0/=100/-0 50.00% 50.0/100 2500.00
2 Torch 2 x64 1CPU +0/=100/-0 50.00% 50.0/100 2500.00

so my idea is: Torch scores better against SF with balanced openings and not so good against weaker engines;
and vice versa with unbalanced openings. This happens with all engines I think. So I usally use for testing 50% stronger and 50% weaker engines - this is not easy against nr. 1 and nr. 2 of the list.