Chess System Tal 1.33 on test

Posts : 1254 Join date : 2020-11-17 Location : France

http://rebel13.nl/a/grl.htm

Subject: Re: Chess System Tal 1.33 on test Sun May 29, 2022 12:59 am

It's busy after the big three. About 6 engines competing for place 4,

Subject: Re: Chess System Tal 1.33 on test Sun May 29, 2022 7:24 am

Estimated CCRL elo : 3553

Code:: Running : gauntlet Chess-System-Tal-NNUE-1.33-2 [average CCRL elo pool 3552]
Time Control : Time control : 40/120
Games : 800

No. Name Win Draw Loss Unf. Score Games %
----------------------------------------------------------------
  1 CST-1.33-2 +148 =508 -144 *0 402.0 800 50.2%
  2 Koivisto 8.0 +37 =134 -29 *0 104.0 200 52.0%
  3 Komodo Dragon +44 =119 -37 *0 103.5 200 51.8%
  4 Berserk 8.5.1 +39 =122 -39 *0 100.0 200 50.0%
  5 Ethereal 13.50-NNUE +24 =133 -43 *0 90.5 200 45.2%

First test I did for Chris was on March 9 with version 1.26

Code:: No. Name Win Draw Loss Unf. Score Games %
-------------------------------------------------------------------------
  1 Chess-System-Tal-NNUE-1.26-v +60 =662 -278 *0 391.0 1000 39.1%
  2 Komodo-Dragon    +177 =307 -16 *0 330.5 500 66.1%
  3 Ethereal 13.50-NNUE +101 =355 -44 *0 278.5 500 55.7%

And compare the progress CSTAL has made against Komodo Dragon and Ethereal 13.50

Code:: Komodo Ethereal
CSTAL 1.26 33.9% 44.3%
CSTAL 1.33 48.2% 54.8%

Posts : 1254 Join date : 2020-11-17 Location : France

Thanks Ed

Subject: Re: Chess System Tal 1.33 on test Tue May 31, 2022 10:17 am

I have started to create a good testing environment for CSTAL and later for REBEL by updating the outdated GRL list in a special way, by lowering the time control to 40/40 as 40/120 became impossible to maintain due to NNUE development for which I need all my computer power. Nowadays current top-10 play so strong it will make little difference and results will remain reliable.

So instead of testing Chris artwork against 4 engines and 800 games it will allows us to play against 12 different opponents and 2400 game matches. A much more reliable way of testing.

A start has been made to accomplish that, SF15 first, result will be posted in the sticky sticky GRL thread.

Total games : 11.200

Each new entry in the GRL will play 2200 games.

New entries:
1. Stockfish 15
2. CSTal-NNUE 133-3
3. Koivisto 8.0
4. Berserk 8.5.1
5. Seer 2.5.0
6. Ethereal 13.50-NNUE
7. RubiChess 20220223

Estimated time : 2-3 days

Preview SF15

Code:: No. Name Win Draw Loss Unf. Score Games %
----------------------------------------------------------------
  1 Stockfish_15 +399 =216 -4 *0 507.0 619 81.9%
  2 Komodo-Dragon_2.5 +2 =32 -23 *0 18.0 57 31.6%
  3 CSTal-NNUE_133-3 +0 =26 -31 *0 13.0 57 22.8%
  4 Koivisto_8.0 +0 =26 -30 *0 13.0 56 23.2%
  5 SlowChess_2.8 +0 =26 -30 *0 13.0 56 23.2%
  6 Seer_2.5.0 +1 =21 -34 *0 11.5 56 20.5%
  7 Berserk_8.5.1 +0 =21 -35 *0 10.5 56 18.8%
  8 RubiChess_20220223 +0 =21 -35 *0 10.5 56 18.8%
  9 Ethereal_13.50-NNUE +1 =13 -42 *0 7.5 56 13.4%
10 Igel_3.0.5 +0 =11 -45 *0 5.5 56 9.8%
11 h6 +0 =11 -46 *0 5.5 57 9.6%
12 Pedone_3.1 +0 =8 -48 *0 4.0 56 7.1%

Subject: Re: Chess System Tal 1.33 on test Wed Jun 01, 2022 10:12 pm

Finished the GRL update, 7 new entries.

http://rebel13.nl/grl-best-40-2.html [best version only]
http://rebel13.nl/grl-40-2.html [all versions]

I renamed the GRL into the "PEACE list" - Private Ed and Chris Engine List as testing environment for CSTAL and later for REBEL.

Used time control is 40/80 to synchronize our different PC's better regarding speed in order that results are in harmony.

Individual results CSTAL

Code:: CSTal-NNUE_133-4 3542.0 : 1400 (+324,=857,-219), 53.8 %

   vs. : games ( +, =, -), (%) : Diff, SD, CFS (%)
   Komodo-Dragon_2 : 200 ( 19, 135, 46), 43.3 : -43.2, 8.8, 0.0
   Koivisto_8.0 : 200 ( 40, 125, 35), 51.3 : -5.4, 9.6, 28.6
   Berserk_8.5.1 : 200 ( 43, 111, 46), 49.3 : -0.3, 12.7, 49.0
   Ethereal_13.50-NNUE : 200 ( 46, 126, 28), 54.5 : +24.6, 8.6, 99.8
   SlowChess_2.8 : 200 ( 47, 129, 24), 55.8 : +58.8, 7.7, 100.0
   RubiChess_20220223 : 200 ( 62, 121, 17), 61.3 : +68.6, 7.9, 100.0
   Seer_2.5.0 : 200 ( 67, 110, 23), 61.0 : +81.0, 10.8, 100.0

Posts : 1254 Join date : 2020-11-17 Location : France

Admin wrote:

Finished the GRL update, 7 new entries.

http://rebel13.nl/grl-best-40-2.html [best version only]
http://rebel13.nl/grl-40-2.html [all versions]

I renamed the GRL into the "PEACE list" - Private Ed and Chris Engine List as testing environment for CSTAL and later for REBEL.

Used time control is 40/80 to synchronize our different PC's better regarding speed in order that results are in harmony.

Individual results CSTAL

Code:: CSTal-NNUE_133-4 3542.0 : 1400 (+324,=857,-219), 53.8 %

   vs. : games ( +, =, -), (%) : Diff, SD, CFS (%)
   Komodo-Dragon_2 : 200 ( 19, 135, 46), 43.3 : -43.2, 8.8, 0.0
   Koivisto_8.0 : 200 ( 40, 125, 35), 51.3 : -5.4, 9.6, 28.6
   Berserk_8.5.1 : 200 ( 43, 111, 46), 49.3 : -0.3, 12.7, 49.0
   Ethereal_13.50-NNUE : 200 ( 46, 126, 28), 54.5 : +24.6, 8.6, 99.8
   SlowChess_2.8 : 200 ( 47, 129, 24), 55.8 : +58.8, 7.7, 100.0
   RubiChess_20220223 : 200 ( 62, 121, 17), 61.3 : +68.6, 7.9, 100.0
   Seer_2.5.0 : 200 ( 67, 110, 23), 61.0 : +81.0, 10.8, 100.0

Thanks Ed

Subject: Re: Chess System Tal 1.33 on test Wed Jun 08, 2022 8:11 am

New development version : CSTal-NNUE 134-E195 +19

Code::    # PLAYER : RATING ERROR POINTS PLAYED (%) CFS(%) W D L D(%)
   1 Stockfish 15 : 3761.7 19.8 1647.0 2000 82 100 1315 664 21 33
   2 Komodo-Dragon 2.5 : 3648.2 8.9 2668.5 4000 67 100 1648 2041 311 51
   3 Lc0 v28 : 3582.8 10.5 591.5 1000 59 95 335 513 152 51
   4 CSTal-NNUE 134-E195 : 3561.0 19.5 789.5 1400 56 94 363 853 184 61
   5 Koivisto 8.0 : 3547.3 7.9 1012.0 2200 46 87 342 1340 518 61
   6 Berserk 8.5.1 : 3541.7 15.8 989.5 2000 49 81 397 1185 418 59
   7 Ethereal 13.50-NNUE : 3520.4 19.7 833.0 1800 46 83 291 1084 425 60
   8 SlowChess 2.8 : 3482.1 8.4 2077.0 4400 47 53 838 2478 1084 56
   9 RubiChess 20220223 : 3473.3 17.1 706.5 1800 39 95 187 1039 574 58
  10 Seer 2.5.0 : 3460.3 14.6 596.5 1600 37 73 167 859 574 54

Congrats Chris [!]

Posts : 1254 Join date : 2020-11-17 Location : France

Thanks Ed. Impressive service! It get built in the afternoon, rated live online overnight, and result for morning coffee!
This current one is about halfway through its training run, so chances are there’s something better, we will see. And, on another PC, a new architecture version is training, hopefully better than this one. Fun!

» Alternative Chess Test 4
» Live Chess 960 Match Dragon by Komodo Chess vs Stockfish 291120 (TC=90m+30s)(32 Threads)
» How strong can the smartest chess engine in the world play intuitive chess?
» Computer Chess Engine Classical Chess Championship. Stockfish 14 vs. Dragon 2.5.1
» Top Chess Engine Fischer Random Rapid Chess Tournament.