Question to Aart Bik

[Nordlandia]

H.G.M: maybe you can start a thread on the winboard/xboard forum about hourglass tc. My account need to be activated and can take some time.

In "Hour Glass" options players start with the same amount of time on the clock (say one minute) but everytime one player's clock is counting down, the other player's clock is counting up.

Hourglass example video: https://youtu.be/9jwhRSziulc

[hgm]

I think you are right that the GUI could simply implement it this way for xboard and also that most engines probably will just use up (part of) the remainder time. I just thought it would be a little cleaner to provide the engine with the exact time control details, just at the off chance some engine author has clever ideas how to exploit hour glass.....

Well, the key phrase here indeed seems to be "off chance". In my estimate it is ~99.9% certain that there will be exactly zero engines that would use any of this ever, if it existed. To make hourglass mode useful in a GUI, the GUI would need to use some work-around to be able to use it with engines that do not support it. And if that work-around already works so well that it is almost impossible to improve on, who would care to improve on it?

[abik]

Well, the key phrase here indeed seems to be "off chance". In my estimate it is ~99.9% certain that there will be exactly zero engines that would use any of this ever, if it existed. To make hourglass mode useful in a GUI, the GUI would need to use some work-around to be able to use it with engines that do not support it. And if that work-around already works so well that it is almost impossible to improve on, who would care to improve on it?

Thanks HG, again, for your insights. All this makes me indeed wonder if hourglass is really that different from e.g. fixed time per time period, but it perhaps makes the game more spectacular to watch...

[Nordlandia]

Thanks HG, again, for your insights. All this makes me indeed wonder if hourglass is really that different from e.g. fixed time per time period, but it perhaps makes the game more spectacular to watch...

Atleast engines don't run out of time. Time trouble is probably non-existent for certain level engines. Recent added adjudication rules is great combination in that regard.

[hgm]

There is a paradox with this hourglass TC that I don't really grasp. Suppose you are playing a ponder-on game with it. Any time you use thinking, is now not only added to the time the opponent is pondering, but also to the time he gets for his next move. Even if you assume ponder time is only 60% efficient, he would still get 1.6 times as much time added to his next move than what you use now. So thinking seems to boost the quality of his moves more than the quality of your own. With as the logical conclusion that you better not think at all...

But that sounds a bit weird.

[Nordlandia]

H.G.Muller: thinking during the opponent's move should not affect clock. Only during each turn. And yes ponder enabled with hourglass TC may produce high-quality chess.

[hgm]

thinking during the opponent's move should not affect clock.

No, but it affects the quality of your move. And in a ponder-on game he also thinks while your clock is running.

[Nordlandia]

Some claims ponder on increase quality of play by about 50 elo under right condidtions, like two identical computers facing each other.

[hgm]

Well, that is to be expected. Rule of thumb is that doubling thinking time increases Elo by ~70 (before draw saturation sets in). Ponder-hit frequencies are typically 60% (and larger in self play), so this would amount to a factor 1.6 in thinking time. With the formula Elo = 100*log(T) that would give 47 Elo.

[Nordlandia]

Well, that is to be expected. Rule of thumb is that doubling thinking time increases Elo by ~70 (before draw saturation sets in). Ponder-hit frequencies are typically 60% (and larger in self play), so this would amount to a factor 1.6 in thinking time. With the formula Elo = 100*log(T) that would give 47 Elo.

Ponder is very rarely used because it's very expensive. You need identical hardware for reliable test results. Uneven hardware generate much noise.

Question: is there any difference between using ponder on dual socket motherboard than two identical desktop machines?

By using dual socket motherboard there is no latency in receiving moves.