Thanks for testing!Wolfgang wrote: ↑Fri Aug 20, 2021 8:00 pm Another update of our 40/4 list is online:
best singles: http://www.cegt.net/40_4_Ratinglist/40_ ... liste.html
all versions: http://www.cegt.net/40_4_Ratinglist/40_ ... liste.html
New engines/versions:
Clover 2.4 x64 1CPU (+116 to v2.3.1)
Seer 2.2.0NN x64 1CPU [only @"all versions" list because of v2.3.0 which is hugely stronger]
Seer 2.3.0NN x64 1CPU (+101 to v2.2.0)
Weiss 2.0 x64 1CPU (+75 to v1.4)
Ethereal 13.25 NNUE x64 1CPU (+52 to v13.00)
DanaSah 9.0NN (DN2) x64 (+63 to v8.8 with DN1)
CEGT-40/4: actual tests
Moderators: hgm, Rebel, chrisw
-
- Posts: 533
- Joined: Sun Sep 06, 2020 4:40 am
- Full name: Connor McMonigle
Re: CEGT-40/4: actual tests
-
- Posts: 347
- Joined: Tue Nov 19, 2019 4:34 am
- Location: https://github.com/TerjeKir/weiss
- Full name: Terje Kirstihagen
Re: CEGT-40/4: actual tests
Thanks for testing Weiss!
-
- Posts: 896
- Joined: Sat May 13, 2006 1:08 am
Re: CEGT-40/4: actual tests
New engines/versions @ 1st 8-Core tourney: https://cegt.forumieren.com/t1511-testi ... gines-8cpu
Second 8-Core tourney started: https://cegt.forumieren.com/t1567-testi ... ament-no-2
LCZero 0.28.0 610000: https://cegt.forumieren.com/t1522-testing-lczero#3034
Bit-Genie 9: https://cegt.forumieren.com/t1568-testing-bit-genie-9
Second 8-Core tourney started: https://cegt.forumieren.com/t1567-testi ... ament-no-2
LCZero 0.28.0 610000: https://cegt.forumieren.com/t1522-testing-lczero#3034
Bit-Genie 9: https://cegt.forumieren.com/t1568-testing-bit-genie-9
-
- Posts: 896
- Joined: Sat May 13, 2006 1:08 am
Re: CEGT-40/4: actual tests
Arasan 23.0.1NN: https://cegt.forumieren.com/t1571-testi ... n-23-0-1nn
Drofa 3.2.0: https://cegt.forumieren.com/t1572-testing-drofa-3-2-0
Zahak 6.0: https://cegt.forumieren.com/t1573-testing-zahak-6-0
all tests with 64bit, 1CPU
Drofa 3.2.0: https://cegt.forumieren.com/t1572-testing-drofa-3-2-0
Zahak 6.0: https://cegt.forumieren.com/t1573-testing-zahak-6-0
all tests with 64bit, 1CPU
-
- Posts: 896
- Joined: Sat May 13, 2006 1:08 am
Re: CEGT-40/4: actual tests
Koivisto 6.16NN: https://cegt.forumieren.com/t1576-testi ... sto-6-16nn
SlowChess Blitz 2.7NN: https://cegt.forumieren.com/t1575-testi ... litz-2-7nn
Counter 3.9: https://cegt.forumieren.com/t1577-testi ... r-3-9#3050
SlowChess Blitz 2.7NN: https://cegt.forumieren.com/t1575-testi ... litz-2-7nn
Counter 3.9: https://cegt.forumieren.com/t1577-testi ... r-3-9#3050
-
- Posts: 896
- Joined: Sat May 13, 2006 1:08 am
Re: CEGT-40/4: actual tests
Komodo Dragon 2.5 (defalut): https://cegt.forumieren.com/t1580-testi ... -mcts#3057 +73 so far
Komodo Dragon 2.5 (MCTS): https://cegt.forumieren.com/t1580-testi ... -mcts#3058 +76 so far
Komodo Dragon 2.5 (MCTS): https://cegt.forumieren.com/t1580-testi ... -mcts#3058 +76 so far
-
- Posts: 671
- Joined: Sun Jan 26, 2020 10:38 pm
- Location: Turkey
- Full name: Mehmet Karaman
Re: CEGT-40/4: actual tests
Impressive results for a top chess engine. It would be nice if Larry could give more details about the newly used NNUE net. In the preparation of this new net , I wonder if Lc0 games were also used .Wolfgang wrote: ↑Sun Sep 26, 2021 8:18 pm Komodo Dragon 2.5 (defalut): https://cegt.forumieren.com/t1580-testi ... -mcts#3057 +73 so far
Komodo Dragon 2.5 (MCTS): https://cegt.forumieren.com/t1580-testi ... -mcts#3058 +76 so far
-
- Posts: 5960
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
Re: CEGT-40/4: actual tests
My estimated elo gain for single thread blitz in the announcement was 72 elo; it is so far at 73 per CEGT, sorry for the one elo error in my forecast! Seriously, if we end up above Stockfish 13 in standard chess and above Stockfish 14 in FRC, my goals for the release will be met. To Answer your question, we trained primarily against Stockfish for this version, but only trained on Komodo's moves and scores, not Stockfish moves or scores. We did not train against Lc0 yet, but you can expect us to do so. We used some ideas from Stockfish, but changed them substantially and probably improved them. Of course we don't use Stockfish code at all. Dietrich Kappe is primarily responsible for training the nets, with Dmitry Pervov supplying the software he needs to train and run them, and with Mark Lefler providing substantial support. My main role is parameter tuning.mehmet123 wrote: ↑Sun Sep 26, 2021 8:36 pmImpressive results for a top chess engine. It would be nice if Larry could give more details about the newly used NNUE net. In the preparation of this new net , I wonder if Lc0 games were also used .Wolfgang wrote: ↑Sun Sep 26, 2021 8:18 pm Komodo Dragon 2.5 (defalut): https://cegt.forumieren.com/t1580-testi ... -mcts#3057 +73 so far
Komodo Dragon 2.5 (MCTS): https://cegt.forumieren.com/t1580-testi ... -mcts#3058 +76 so far
Komodo rules!
-
- Posts: 896
- Joined: Sat May 13, 2006 1:08 am
-
- Posts: 6993
- Joined: Thu Aug 18, 2011 12:04 pm
Re: CEGT-40/4: actual tests
Nice service.Wolfgang wrote: ↑Sat Oct 02, 2021 6:44 pm Colossus 2021a: https://cegt.forumieren.com/t1581-testi ... ssus-2021a
Benjamin 1.1: https://cegt.forumieren.com/t1587-testi ... n-1-1#3072
In case you have own server space I can make my util available that updates the progression of matches automatically, examples :
http://rebel13.nl/a/grl.htm
http://rebel13.nl/b/grl.htm
Offer is for anyone interested.
90% of coding is debugging, the other 10% is writing bugs.