Fastest Time Controls for CuteChess-Cli?

[hgm]

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

There always is a limit. The 1-1.5 msec communication delay I quoted applied just as much to Linux as to Windows, and if you would give the engine less than 1 msec/move, it could not keep up and should lose on time.

What is worse, most engines measure the time they use on a system clock that ticks only 50 or 60 times per second. So they start thinking, and upto the first tick, they will always think: "wow, I am doing great, only used 0 msec so far, plenty of time to go on". And then the clock ticks, and they think: "Oops, 16 msec used. And I only had 5 msec left on my clock! Must move immediately!". And what is worse, the GUI will think: this engine had only 5 msec left, and I receive its move only after 11 msec, so let's give it a nice forfeit!

Linux clock calls might promise msec precision, but they still increase in steps of 16. Precision is not the same as accuracy...

[tpetzke]

Hi Michael,

my experience tells me the opposite. Normally the effect on elo is
smaller with longer time controlls

this is what I actually meant. A promising improvement a shorter TC will have a smaller effect at larger ones but will probably still be an improvement.

An eval term for a trapped bishop might be very valuable for short TCs because the low search depths put the loss of the bishop behind the horizon. At longer TCs the horizon is further away and search will help to prevent some of those losses.

So we are at the same page. Sorry if I expressed myself unclear.

Thomas...