Fastest Time Controls for CuteChess-Cli?

[Steve Maughan]

What are the fastest time controls you use with CuteChess-Cli? I'm doing 40 moves in 15 seconds and occasionally get a message:

Warning: QObject:: killTimers: timers cannot be stopped from another thread

I'm not sure if this is due to overload or it's a problem with Maverick.

Does anyone know if this is a problem or just FYI?

Thanks,

Steve

[Adam Hair]

I thought that Ilari gave an explanation for that warning, but I can not find the post. Regardless, it does not seem to be a problem.

As for the time control, you can probably go as fast as Maverick can go. I have used 40 moves / 4 seconds and less without problem with some engines.

[hgm]

I once created a testing tool for GUI timing (it should still be on my website somewhere), and used it to time the engine-to-engine delay of cutechess-cli and WinBoard / XBoard under various conditions. On a 1.3GHz Celeron M the delay was typically 1-1.5 msec. So it seems that if you stay well above 1 msec/move, you should have no problems, neither with cutechess-cli nor with xboard -noGUI.

On faster machines it might be possible to go even faster, but you would have to perform the timing measurement yourself there to be sure.

[Ajedrecista]

Hello Steve:

What are the fastest time controls you use with CuteChess-Cli? I'm doing 40 moves in 15 seconds and occasionally get a message:

Warning: QObject:: killTimers: timers cannot be stopped from another thread

I'm not sure if this is due to overload or it's a problem with Maverick.

Does anyone know if this is a problem or just FYI?

Thanks,

Steve

I tried to find Ilari's explanation that Adam refers to... I think I have not find it. Anyway, here is a thread where this warning is briefly reported:

Cutechess-cli 0.4.0 released

FWIW, I did a fixed depth testing a year ago in a try to quantify the effect of diminishing returns. You will see that this warning sometimes appeared in my tests at the end of them without causing problems:

Diminishing returns in fixed depth testing revisited.

Good luck with your tests!

Regards from Spain.

Ajedrecista.

[nkg114mc]

Hi Steven,

I am very interested to your engine re-writing and also following your blog. Here is just a quick question: how do you decide what time control to use in your testing? I know that there are variety of setting for different type of testing, but, like such kind of fix-time testing, it is obvious related to the computational resources you have. For example, I have only one desktop computer, a short time control (like 15 seconds / 40 moves) is helpful to get an elo score soon when finishing playing a large number of games (say 8000+). Let's say we testing two engines A, and B after 8000 games, and A's elo is higher than B. Is this conclusion still solid for some other time control (e.g. 40 min / 40 moves)?

[lucasart]

What are the fastest time controls you use with CuteChess-Cli? I'm doing 40 moves in 15 seconds and occasionally get a message:

Warning: QObject:: killTimers: timers cannot be stopped from another thread

I'm not sure if this is due to overload or it's a problem with Maverick.

Does anyone know if this is a problem or just FYI?

Thanks,

Steve

There is NO limit.

I have played tens of thousands of games in 2.5"+0.025", using 7 concurrent games (on a 8 core machine). Not a single time loss.

Here is the secret: I don't use Windows

[tpetzke]

Hi Ma,

whether you can use a short TC test result as given also for larger TC depends on what you test.

Obviously not if you test changes in the time control behavior. Also if you test features that require a larger search depth it doesn't scale well.

King safety is often mentioned as a feature that requires a bit of a depth to show its value.

But for the majority of features an improvement at short TC will also work at longer ones, but maybe its effect is much smaller then.

Thomas...

[tpetzke]

Here is the secret: I don't use Windows

I do, and I don't have time losses either, at least on the side of my engine, even when running 4 concurrent games on my 4 core machine.

It is easy to blame the OS, but usually in my tests it was the engine and sometimes it was the GUI.

Thomas...

[nkg114mc]

Thanks a lot for the answer, Thomas! This makes sense to me.

[Desperado]

Hi Ma,

... but maybe its effect is much smaller then.

Thomas...

Hello Thomas,

my experience tells me the opposite. Normally the effect on elo is
smaller with longer time controlls ( in the majority of test cases ).

In addition this observation fits into other test scenarios like depth tests,
where in many cases a 20 elo gain drops under 5 elo as example, the deeper you search.

Search seems to assimilate knowledge somehow the deeper you search,
and progress of a deeper search gets another depth ratio.
My impression is there is a correlation between depth progress like going to 6 from 5 ( ratio 5/6 ) and for example 12/13.

Time based games ( lightspeed games ) follow at least a similar behavior.

Michael