TT behavior

Discussion of chess software programming and technical issues.

Moderators: hgm, Rebel, chrisw

Karlo Bala
Posts: 373
Joined: Wed Mar 22, 2006 10:17 am
Location: Novi Sad, Serbia
Full name: Karlo Balla

TT behavior

Post by Karlo Bala »

A Long time ago I had a classical TT and later I decided to put only fail high scores in a TT. Recently I changed scheme to save both, fail high and fail low scores. To my big surprise, it turns out that in most endgames search is getting more performance from fail low scores from TT then from fail high scores. I have no good explanation. Does anyone have?
Best Regards,
Karlo Balla Jr.
bob
Posts: 20943
Joined: Mon Feb 27, 2006 7:30 pm
Location: Birmingham, AL

Re: TT behavior

Post by bob »

Karlo Bala wrote:A Long time ago I had a classical TT and later I decided to put only fail high scores in a TT. Recently I changed scheme to save both, fail high and fail low scores. To my big surprise, it turns out that in most endgames search is getting more performance from fail low scores from TT then from fail high scores. I have no good explanation. Does anyone have?
Pretty simple idea. Fail high nodes require you to search just one move, typically, to get the fail high. Low-effort. When you get a TT hit, and the draft/bound is acceptable, you save that low effort search. Fail-low nodes require you to search _every_ move, which is much higher effort. A TT hit saves that high-effort search...

Ditto for PV hits (EXACT), except that there are just not as many of them.