CEGT - rating lists June 21th 2020

Discussion of computer chess matches and engine tournaments.

Moderators: Harvey Williamson, bob, hgm

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
User avatar
Werner
Posts: 2547
Joined: Wed Mar 08, 2006 9:09 pm

CEGT - rating lists June 21th 2020

Post by Werner » Sun Jun 21, 2020 5:33 pm

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

40 / 20:
New games: 1.924; 23 different engines
Total:   1.356.178

NEW Engines
2 LCZero 0.25.1 Cuda (LS15.0): 3514 - 499 games (good start +20 to LS14)
351 Combusken 1.2.0 x64 1CPU: 2954 - 700 games (-)
404 Marvin 3.6.0 x64 1CPU: 2919 - 700 games (+68 to v. 3.5.0)

UPDATES
333 Winter 0.8 x64 1CPU: 2965 - 1307 games (+1)
1 Stockfish 11.0 x64 8CPU: 3515 - 900 games (-3)

40 / 4
New games: 13.086
Total now: 2.760.062

New engines
232 Arasan 22.0 x64 8CPU: 3213 - 500 games (+22 to v. 21.4)
150 Defenchess 2.2 x64 8CPU: 3291 - 500 games (+126 to v. 2.0)
169 Fritz 17 x64 8CPU: 3272 - 500 games (-)
116 RofChade 2.3 x64 8CPU: 3328 - 800 games (+21 to v. 2.2)
194 Stockfish DD x64 8CPU: 3245 - 500 games (-)
198 Booot 6.4 x64 2CPU: 3241 - 700 games (+53 to v. 6.3)
238 Defenchess 2.2 x64 2CPU: 3208 - 700 games (-)
108 Ethereal 12.00 x64 2CPU: 3336 - 1200 games (+35 to v. 11.75)
94 Komodo 14.0 x64 2CPU (MCTS): 3352 - 700 games (+30 to v. 13.2.5)
261 RubiChess 1.7.1 x64 2CPU: 3187 - 700 games (-)
211 SlowChess Blitz Classic 2.1 2CPU: 3231 - 700 games (-)

9 LCZero 0.25.1 Cuda (LS14.0): 3536 - 400 games (+53 to v. LS13)
309 RubiChess 1.7.3 x64 1CPU: 3142 - 1500 games (+16 to v. 1.7.1)
2523 Sargon 1978 1.00 UCI: 1092 - 900 games (-)
621 Winter 0.8 x64 1CPU: 2947 - 1200 games (+112 to v. 0.7)

Updates
297 Equinox 3.30 x64 8CPU: 3152 - 950 games (-20)
181 Fizbo 2.0 x64 8CPU: 3258 - 1140 games (-2)
221 Gull 3.0 x64 8CPU: 3219 - 1950 games (-1)
139 Laser 1.7 x64 8CPU: 3302 - 1340 games (+3)
288 Nirvana Chess 2.4 x64 8CPU: 3158 - 950 games (-3)
125 Schooner 2.2 x64 8CPU: 3311 - 1600 games (+8)
29 Komodo 14.0 x64 4CPU: 3472 - 1136 games (-12)
50 Komodo 14.0 x64 2CPU: 3425 - 1200 games (+-0)
178 RofChade 2.3 x64 2CPU: 3261 - 1600 games (+8)
147 Xiphos 0.6 x64 2CPU: 3295 - 2900 games (+4)
278 SlowChess Blitz Classsic 2.1: 3167 - 2000 games (-3)
12 Stockfish 20200530 x64 1CPU: 3526 - 1600 games (+3)

40 / 120:
Last update was on 17th of June! this list is now closed.

We started a new list 25´+8´´ instead : http://www.cegt.net/25plus8Rating/Purel ... liste.html
  • no Program Elo + - Games Score Av.Op. Draws
    1 LCZero 0.25.1 Cuda (LS15.0) 3506 44 44 200 61.5% 3422 71.0%
    2 Stockfish 11.0 x64 3448 19 19 700 67.9% 3306 55.7%
    3 Stoofvlees II a14 3395 64 64 100 35.0% 3506 70.0%
    4 Komodo 14.0 x64 3352 20 20 700 60.7% 3270 63.4%
    5 Houdini 6.0 x64 3334 22 22 700 58.1% 3272 60.9%
    6 Komodo 14.0 x64 (MCTS) 3292 22 22 600 51.8% 3279 62.8%
    7 Ethereal 12.25 x64 3274 22 22 600 49.4% 3279 62.8%
    8 RofChade 2.3 x64 3200 22 22 700 38.2% 3291 58.1%
    9 Booot 6.4 x64 3187 23 23 600 33.3% 3317 54.7%
    10 Fritz 17 x64 3154 26 26 500 32.1% 3290 53.8%
5'+3'' pb=on
Last update was June 08th
Komodo 14.0 x64 3367/1600 -3
Komodo 14.0 x64 (MCTS) version 3295/1700 +9
LCZero 0.25.1 Cuda (LS14.0) 3499/800

3'+1'' pb=on
Last update was June 16th
with Komodo 14.0 x64 3371/2800 +2
Komodo 14.0 x64 (MCTS) version 3280/2800 +12

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 Schüle
CEGT-Team

lkaufman
Posts: 4339
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: CEGT - rating lists June 21th 2020

Post by lkaufman » Tue Jun 23, 2020 3:38 am

Congratulations on your decision to start the 25' + 8" list. I hope this will become the most significant of your lists. It has several desirable properties:
1. Use of increment; repeating time controls are a waste of resources
2. no ponder; while some may care about ponder, it is also a waste of resources, and not relevant for people who use engines for analysis.
3. non-blitz time control, more in the range of normal use I would say.
4. Higher base to increment ratio; appropriate for non-blitz games. Something like 200 to 1 (you are using 187.5 to 1) is reasonable for longer games, based on some tests I ran a while back. More than 200 to 1 is probably not a good idea.
5. clean start means the relative ratings should be pretty accurate

Three questions:
1. Why exactly 25' + 8"? It is quite reasonable, but I've never heard of a single game being played at that precise time limit. Perhaps the answer has to do with my next question:
2. Is this the actual time limit used on the I7 machines, or do you divide by some constant like 3 or so to get the actual time limit adjusted from ancient to modern hardware? If so, what is the actual time limit used on the I7 machines?
3. Will this list be limited to single-core only testing?
Komodo rules!

ThatsIt
Posts: 971
Joined: Thu Mar 09, 2006 1:11 pm
Contact:

Re: CEGT - rating lists June 21th 2020

Post by ThatsIt » Tue Jun 23, 2020 7:01 am

Hi Larry,

we've chosen 25'+8" because it's nearly in between the fastGMs lists 10'+6" and 60'+15".

We do not apply any time adjustment as you can see here:
http://www.cegt.net/rating25plus8.htm
We hope that our hardware will be consisted for the next few years ;-)

Any yes, this list will be "1 core only".

Best wishes,
G.S.
(CEGT team)

lkaufman
Posts: 4339
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: CEGT - rating lists June 21th 2020

Post by lkaufman » Tue Jun 23, 2020 5:57 pm

ThatsIt wrote:
Tue Jun 23, 2020 7:01 am
Hi Larry,

we've chosen 25'+8" because it's nearly in between the fastGMs lists 10'+6" and 60'+15".

We do not apply any time adjustment as you can see here:
http://www.cegt.net/rating25plus8.htm
We hope that our hardware will be consisted for the next few years ;-)

Any yes, this list will be "1 core only".

Best wishes,
G.S.
(CEGT team)
OK, thanks. I missed the "no time adjustment" line. Do you anticipate that a lot of your testing resources will go into this list? If it is likely to become one of your most important lists (meaning lots of games, new versions tested, etc.) we might reconsider the base to increment ratio we use in our testing.
Komodo rules!

ThatsIt
Posts: 971
Joined: Thu Mar 09, 2006 1:11 pm
Contact:

Re: CEGT - rating lists June 21th 2020

Post by ThatsIt » Wed Jun 24, 2020 7:39 am

Hi Larry,

the NN-testing is the only problem because the whole machine is blocked for a 100-games-match. That takes time. A/B-engines-testing is no problem, with "concurrency 5" 100 games are ready in about 27-28 hours.
Our 40/120 list was much more complex in terms of time consumption.

Best wishes,
G.S.
(CEGT team)

Modern Times
Posts: 2604
Joined: Thu Jun 07, 2012 9:02 pm

Re: CEGT - rating lists June 21th 2020

Post by Modern Times » Wed Jun 24, 2020 7:57 pm

ThatsIt wrote:
Wed Jun 24, 2020 7:39 am
Hi Larry,

the NN-testing is the only problem because the whole machine is blocked for a 100-games-match. That takes time. A/B-engines-testing is no problem, with "concurrency 5" 100 games are ready in about 27-28 hours.

Best wishes,
G.S.
(CEGT team)
Yes painful. But on a 6-core machine for example, could you say run the NN match in Arena, and then fire up cutechess and run A/B engines in that with concurrency to take you up to 5 cores ? So running two GUIs simultaneously ?
.

Opinions expressed here are my own, and not necessarily those of the CCRL Group.

Wolfgang
Posts: 354
Joined: Fri May 12, 2006 11:08 pm

Re: CEGT - rating lists June 21th 2020

Post by Wolfgang » Wed Jun 24, 2020 9:07 pm

Modern Times wrote:
Wed Jun 24, 2020 7:57 pm
....
Yes painful. But on a 6-core machine for example, could you say run the NN match in Arena, and then fire up cutechess and run A/B engines in that with concurrency to take you up to 5 cores ? So running two GUIs simultaneously ?
Yes, that's possible and normally causes no problems. So it's good for the rating list, because we can use the machine for other matches simultaneously, but it doesn't shorten the NN-match (only one instance possible), so that these matches take much more time than A/B matches.
Best
Wolfgang
CEGT-Team

Post Reply