Smallest Increment to Not Lose on Time?

[Steve Maughan]

Question: what the smallest move increment to realistically except an engine to never lose on time? Is 10 ms reasonable? Are there any engines that play a 1 second + 10 ms increment and never lose on time?

— Steve

[Jouni]

Just one try with 1s and 10ms.

Code: Select all

Score of stockfish17 vs obsidian150: 212 - 275 - 513 [0.469]
...      stockfish17 playing White: 131 - 113 - 256  [0.518] 500
...      stockfish17 playing Black: 81 - 162 - 257  [0.419] 500
...      White vs Black: 293 - 194 - 513  [0.549] 1000
Elo difference: -21.9 +/- 15.0, LOS: 0.2 %, DrawRatio: 51.3 %
1000 of 1000 games finished.

No time losses. Both have 10ms Move Overhead default.

[Steve Maughan]

Just one try with 1s and 10ms.

Code: Select all

Score of stockfish17 vs obsidian150: 212 - 275 - 513 [0.469]
...      stockfish17 playing White: 131 - 113 - 256  [0.518] 500
...      stockfish17 playing Black: 81 - 162 - 257  [0.419] 500
...      White vs Black: 293 - 194 - 513  [0.549] 1000
Elo difference: -21.9 +/- 15.0, LOS: 0.2 %, DrawRatio: 51.3 %
1000 of 1000 games finished.

No time losses. Both have 10ms Move Overhead default.

Thanks — Quite amazing. So can I assume 10 ms is the minimum increment, or is 5 ms also possible?

— Steve

P.S. what did you use to run the match — CuteChessCL or FastChess?

[Ajedrecista]

Hello Steve:

Here is a probably unrelated topic or maybe useful, I do not know. Just in case:

Win 7 GetTickCount() weirdness

Please note that thread is from 2013, so long time ago. Then, people wrote about 1/64th of second (15.625 ms) for some timer resolutions and that WB protocol passed time control info to the engines in bunches of 10 ms.

Good luck with your research.

Regards from Spain.

Ajedrecista.

[Steve Maughan]

Ajedrecista — great point. Yes, you need a high resolution timer to make an increment of 10 ms possible.

— Steve

[Brunetti]

So can I assume 10 ms is the minimum increment, or is 5 ms also possible?

10 is an arbitrary value; maybe 5 is possible or even 0, who knows? it depends on the engine(s), the time control and the tournament manager...

Alex

[chesskobra]

In cutechess GUI I think it is not possible to set increment less than 1s (but of course 0 is possible). I am interested in a slightly more general question - at 1s + 10ms or even smaller increment, which engines play at high level (in human standards)? For example, will SF playing at 1s +10ms beat Carlsen playing at classical time control?

[Jouni]

I tested in Cute Chess GUI. And there 10 ms is the smallest increment. In log I see both engines have many 1 ms moves "77. Kg6 {+1.97/10 0.001s} "! Of course all testing below 60 + 0.6 is almost useless.

[jefk]

imo a relevant, even interesting thread, while with slower time controls, more and
more games are ending in draws (unless with silly force openings as in Tcec), fast
bullet may (one of the) the way(s) to go.

Some tech -and program specifications- may change for such purposes;
small point: if you get a ponder hit, then an instantaneous move ofcourse
can be very fast; how fast in ms (900 microseconds maybe , i dont know;
and then there's the issue of GUI, internet transmission and stuff as lag ofcourse.