Incremental vs repeating TC

[Frank Quisinsky]

Optimal is:
All programmer used the same clear-code / sources for ...

- time management
- endgame bases support in combination with mate-distance (syzygy)
- ponder
- contempt
- uci support with same possibilities (in additional programmers can set own settings if they like)
- same code for smp, more-processor support

That is indeed best situation for testing engines.
To clone best sources make sense here.

In the past I had some discussion with winboard programmers about it.
But most are thinking ... no, no Frank ... each programmer should do that in own home-working.

Is that today "up-to-date" or better "modern-thinking"?

[AndrewGrant]

The only reason I haven't switched to Fischer time controls is that it would mean abandoning everything I've done and starting again. I'll just continue on with repeating time control until I cease ratings list work completely.

Could be a chance for a new rating list at CCRL. 2+1? Always using 8 cores on engines? Make the list a pure list, where only one copy of each engine impacts the elo tables? Get a resonable GPU for those engines, something that multiple testers have, to have some uniformity.

[RubiChess]

But maybe I should kill all engines with a bad time-mangament, engines where I can't set contempt = 0 or engines produced "lost on time" games with "ponder = on". The question is how many from TOP-41 engines are end of the day in my tourney.

Means end of the day I have only:

Frank Quisinsky 1
Frank Quisinsky 2
Frank Quisinsky 3
and no engines!

System: AMD Ryzen 7 3700X 8-Core Processor with HT enabled
GUI: cutechess-cli
TC: 5.0+0.1 (very short for trying to produce some time forfeits)
Ponder: on
Hash: 64
Threads: 1
Concurrency: 8 (even GUI process doesn't get his own core)
Tablebase: no
Adjudication:
-resign movecount=3 score=600
-draw movenumber=34 movecount=8 score=5

Code: Select all

  1 Stockfish-12                   : 2529   51  49   100    65.5 %   2418   47.0 %
  2 Berserk-9                      : 2525   49  48   100    65.0 %   2418   50.0 %
  3 Ethereal-13.75                 : 2488   47  46   100    60.0 %   2418   54.0 %
  4 Koivisto-8.13                  : 2456   44  43   100    55.5 %   2418   59.0 %
  5 RubiChess-20220626             : 2416   16  16  1000    54.2 %   2386   47.8 %
  6 Stockfish-Classic              : 2390   54  54   100    46.0 %   2418   38.0 %
  7 Slow-2.9                       : 2386   51  51   100    45.5 %   2418   45.0 %
  8 Rofchade-3.0                   : 2336   47  48   100    38.5 %   2418   51.0 %
  9 Seer-2.5                       : 2318   45  47   100    36.0 %   2418   54.0 %
 10 Nemorino-6.0                   : 2227   49  53   100    25.0 %   2418   44.0 %
 11 Igel-3.1.0                     : 2188   55  59   100    21.0 %   2418   36.0 %

Engines were not selected for this ponder test, just used my usual testing set with short TC and ponder on.
The result of this gauntlet is quite unimportant but...
Number of time forfeits: 0

[AndrewGrant]

But maybe I should kill all engines with a bad time-mangament, engines where I can't set contempt = 0 or engines produced "lost on time" games with "ponder = on". The question is how many from TOP-41 engines are end of the day in my tourney.

Means end of the day I have only:

Frank Quisinsky 1
Frank Quisinsky 2
Frank Quisinsky 3
and no engines!

System: AMD Ryzen 7 3700X 8-Core Processor with HT enabled
GUI: cutechess-cli
TC: 5.0+0.1 (very short for trying to produce some time forfeits)
Ponder: on
Hash: 64
Threads: 1
Concurrency: 8 (even GUI process doesn't get his own core)
Tablebase: no
Adjudication:
-resign movecount=3 score=600
-draw movenumber=34 movecount=8 score=5

Code: Select all

  1 Stockfish-12                   : 2529   51  49   100    65.5 %   2418   47.0 %
  2 Berserk-9                      : 2525   49  48   100    65.0 %   2418   50.0 %
  3 Ethereal-13.75                 : 2488   47  46   100    60.0 %   2418   54.0 %
  4 Koivisto-8.13                  : 2456   44  43   100    55.5 %   2418   59.0 %
  5 RubiChess-20220626             : 2416   16  16  1000    54.2 %   2386   47.8 %
  6 Stockfish-Classic              : 2390   54  54   100    46.0 %   2418   38.0 %
  7 Slow-2.9                       : 2386   51  51   100    45.5 %   2418   45.0 %
  8 Rofchade-3.0                   : 2336   47  48   100    38.5 %   2418   51.0 %
  9 Seer-2.5                       : 2318   45  47   100    36.0 %   2418   54.0 %
 10 Nemorino-6.0                   : 2227   49  53   100    25.0 %   2418   44.0 %
 11 Igel-3.1.0                     : 2188   55  59   100    21.0 %   2418   36.0 %

Engines were not selected for this ponder test, just used my usual testing set with short TC and ponder on.
The result of this gauntlet is quite unimportant but...
Number of time forfeits: 0

Anyone who finds time losses -- with any of the OpenBench engines -- is at fault.

[Modern Times]

The only reason I haven't switched to Fischer time controls is that it would mean abandoning everything I've done and starting again. I'll just continue on with repeating time control until I cease ratings list work completely.

Could be a chance for a new rating list at CCRL. 2+1? Always using 8 cores on engines? Make the list a pure list, where only one copy of each engine impacts the elo tables? Get a reasonable GPU for those engines, something that multiple testers have, to have some uniformity.

We can't play enough games with the lists we have, adding a new one is pretty much out of the question.

I have however thought of archiving the existing FRC list and starting again with 2+1, and as you say a new version replaces its predecessor so there is only ever the latest version. However I see no point in 8CPU, you just add games 8 times more slowly than 1CPU. Playing sufficient volume is more important. And that is problematic for GPU engines as there is no concurrency with the GPU like the CPU. With them, assuming you normally run concurrency of 12 with CPU engines, well they take 12 times as long. And anyway, too much controversy these days around chess engines, so retirement from ratings lists isn't too far away for me.

Anyway we are polluting the Komodo thread !

[Frank Quisinsky]

To RubiChess:

From this group of engines, I think no engines will produce time losses.
Two engines have not an optimal time management.

Can be see with the tool "somu"!

Best
Frank

[Frank Quisinsky]

Modern Times:
Welcome in the club (your last sentence)!

Different younger peoples like to go in opposition.
Such a big children's garden never computer chess produced in the past.

8 cores for 2+1



Factor 5.6
2.4 processor units go to nirwana in times energy is expensive.

I have a better idea:
Game in 1 minute + 10 seconds with ponder = on and contempt = 100, of course with hyperthreading = on and own brain = off.

Best
Frank

[RubiChess]

From this group of engines, I think no engines will produce time losses.

I know. This was just to demonstrate the quality of your post regarding "kill all engines with a bad time-mangament... end of the day I have ... no engines".

Regards, Andreas

[Frank Quisinsky]

Hi Andreas,

quality of postings ...

Oh, you like to set again an attack.
Stopped counting since a while.

Yes, what I wrote is right!
The quantity of engines is clealy lesser if I am looking on time mangament, engine errors, contempt problems with long draw games ...

For a while we had a discussion about your RubiChess in CSS forum.
You changed different things and move-average is clearly improved.

To my quality of postings and your stupid comment again.

Maybe next week in an other thread you will get the next chance!



Best
Frank

PS: Different other programmers will like your comment but you need a bit more for a membership in "children programmer's club". Give Gummi!!

[RubiChess]

Frank,

it is YOU who starts ranting about programmers and their engines not satisfying your (very subjective) needings again and again and again. Low move avarage stats and all that blabla...
YOUR stats are the absolute truth, every different opionion is wrong by definition. That exactly is what I learn from your posts since I started to read them.
Everybody who dares to criticise your opionion is an attacker by YOUR definition.
Even proving that you are just not saying the truth (what I did in my post) is an attack or a stupid comment by YOUR definition, cause YOU are right by definition. That is my impression of you since I read chess forums.

No need to give me "the next chance" (shall I say "her majesty"?), I'm done with you.
Sorry to the Komodo team for continuing this hijack, it stops now, at least from my side.

Regards, Andreas

PS. You are talking about engines wasting electricity by bad move stats. Remember: Every single character of your endless and countless posts will consume electricity until the end of time or internet.