CEGT - rating lists March 28th 2021

[Werner]

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

40 / 20:
New games: 2.508; 28 different engines
Total: 1.453.524

NEW Engines
505 Amoeba 3.3 x64 1CPU: 2903 - 1000 games (+21 to v. 3.2)
147 Booot 6.5 x64 1CPU: 3212 - 1108 games (+38 to v. 6.4)
134 Lc0 0.27.0 dnnl 703810: 3230 - 400 games (+39 to v. 0.26.0 )

UPDATES
61 KDragon 1.0 x64 4CPU (MCTS): 3364 - 300 games (startrating)

40 / 4
last update was February 15th: with 8802 new games; now 2.859.072 games
we are testing:
Weiss 1.3 x64 1CPU 491,0/900 games
Halogen 10 = ca. ELO 3052 out of 1100 games (+85 to v9.0)
Texel 1.08a13 x64 1CPU Perf= ~ 2977 out of 1200 games (+4!! to v1.07...)
Counter 3.7 x64 1CPU = ca. ELO 2871 out of 1100 games
Stockfish 13.0 NNUE x64 1CPU = ca. ELO 3620 out of 1400 games (+17 / +39)
Amoeba 3.3 x64 1CPU 2896 out of 1000 games (+17 to v3.2)
Booot 6.5 x64 1CPU 3243 out of 1100 games (+42 to v6.4)
https://cegt.forumieren.com/t1441-testi ... tournament
https://cegt.forumieren.com/t1459-testi ... tournament

25'+8''
last update was March 08th with 1700 new games; total now 29800 games
New engines
we are testing
Booot 6.5 x64
https://cegt.forumieren.com/t1465-for-t ... urney-no-1

5'+3'' pb=on
last update was February 22th with +9000 games
and testing: https://cegt.forumieren.com/t1465-for-t ... urney-no-1

3'+1'' pb=on
Last update was March 3rd - see extra posting.
we are testing: https://cegt.forumieren.com/t1188-for-t ... sions-list

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]

Did the testing of Komodo Dragon MCTS (on any lists) use the AVX2 version or not? Your results for this particular version are way below our own test results, but we use AVX2 so this might be the reason; AVX2 makes a huge difference with Dragon, unlike the case with normal Komodo.

[Wolfgang]

AVX2 of course on all my computers, what else

I have ~ +40 in my 40/4 test:
https://cegt.forumieren.com/t1394-testi ... -mcts#2827

Not enough?

[lkaufman]

AVX2 of course on all my computers, what else

I have ~ +40 in my 40/4 test:
https://cegt.forumieren.com/t1394-testi ... -mcts#2827

Not enough?

Indeed, not nearly enough; we get nearly similar gains from k14.1 to Dragon on both standard mode and MCTS mode, which would mean at least a hundred elo more than you are getting. We don't usually test at repeating time controls, and don't use the range of opponents that you do, but these details normally won't swing ratings by even twenty elo, let alone one hundred. I'll try testing more like the way you do, to see what the explanation might be. This is pretty important to us, as the MCTS mode is not very useful if the gap from standard mode is too large. MCTS has real advantages, but it can't overcome a gap approaching 200 elo if that is the reality.

[pohl4711]

AVX2 of course on all my computers, what else

I have ~ +40 in my 40/4 test:
https://cegt.forumieren.com/t1394-testi ... -mcts#2827

Not enough?

Indeed, not nearly enough;

Yes, this result seems strange. In my ratinglist (sadly my website is still offline), I got this for MCTS (K14 / Dragon):

15 KomodoDragon 1.0 MCTS : 3479 6 6 7000 57.6 % 3425 57.1 %
33 Komodo 14 MCTS : 3338 7 7 5000 44.4 % 3383 53.4 %

Which is +141 Elo.

[lkaufman]

AVX2 of course on all my computers, what else

I have ~ +40 in my 40/4 test:
https://cegt.forumieren.com/t1394-testi ... -mcts#2827

Not enough?

Indeed, not nearly enough;

Yes, this result seems strange. In my ratinglist (sadly my website is still offline), I got this for MCTS (K14 / Dragon):

15 KomodoDragon 1.0 MCTS : 3479 6 6 7000 57.6 % 3425 57.1 %
33 Komodo 14 MCTS : 3338 7 7 5000 44.4 % 3383 53.4 %

Which is +141 Elo.

Overnight I ran KomodoDragon 1.0 MCTS (AVX2) vs Stockfish 11 at 40 moves in 2 minutes repeating, using a fairly normal opening book, and got a result of just minus five elo after over 17,000 games. This would give it 3470 on the CEGT blitz list, 168 elo over Komodo 14.1 MCTS, even more than your excellent +141 elo result (vs K14 mcts). My result is against just one, nearly equal, opponent, so it's not quite the same test, but it should be a pretty fair rating. A result of just +40 elo would seem to be impossibly far below these other two results; just why is the mystery now.

[Werner]

My result for 40/20 list was:
KDragon 1.0 x64 1CPU (MCTS) (3327) - Stockfish 11.0 x64 1CPU (3442) ; performance 3358 = -84.
I have used the openings called TCEC low draw here. Next week I will start the same match with a very balanced opening set from Frank.

[lkaufman]

My result for 40/20 list was:
KDragon 1.0 x64 1CPU (MCTS) (3327) - Stockfish 11.0 x64 1CPU (3442) ; performance 3358 = -84.
I have used the openings called TCEC low draw here. Next week I will start the same match with a very balanced opening set from Frank.

I ran KDragon 1.0 MCTS vs SF11 at 2' + 1" on four threads each overnight on a normal book; result was just -24 elo after 8000 games. I'll rerun on just one thread next.

[lkaufman]

My result for 40/20 list was:
KDragon 1.0 x64 1CPU (MCTS) (3327) - Stockfish 11.0 x64 1CPU (3442) ; performance 3358 = -84.
I have used the openings called TCEC low draw here. Next week I will start the same match with a very balanced opening set from Frank.

On one thread at 2' + 1", again using normal opening book, I got just minus five elo for this pairing after 1900 games, the same result I got at 40/2 min repeating. So the type of time control (increment vs. repeating) doesn't seem to be a significant factor in this puzzle. Now I'm trying five times longer tc, 10' + 5", to see if there is a scaling issue; this should roughly approximate your 40/20 TC adapted to i7.

[lkaufman]

My result for 40/20 list was:
KDragon 1.0 x64 1CPU (MCTS) (3327) - Stockfish 11.0 x64 1CPU (3442) ; performance 3358 = -84.
I have used the openings called TCEC low draw here. Next week I will start the same match with a very balanced opening set from Frank.

On one thread at 2' + 1", again using normal opening book, I got just minus five elo for this pairing after 1900 games, the same result I got at 40/2 min repeating. So the type of time control (increment vs. repeating) doesn't seem to be a significant factor in this puzzle. Now I'm trying five times longer tc, 10' + 5", to see if there is a scaling issue; this should roughly approximate your 40/20 TC adapted to i7.

At 10' + 5" on one thread for above test I got minus three elo after 684 games, so scaling doesn't appear to be a problem either. Since the results are so close to even I wouldn't expect that a low-draw book would make much difference in the elo gap, but I may try it anyway to see.