Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Endgame tablebase format

Author: Bruce Moreland

Date: 10:59:15 06/14/98

Go up one level in this thread



On June 14, 1998 at 03:31:12, jonathan Baxter wrote:

>I find the file probes to be pretty expensive---in KnightCap the disk
>seeks
>can cost up to 2/3 of the processing time. But I have two reasons for
>doing this: speed is one and the other is if you can do it accurately
>then you may be able to generalize to more pieces without having to
>compute all positions.

My endgame database format is probably quite similar to Steven's.  I
hear that he stores everything as one byte, I store my values using the
number of bits necessary to store the largest value in the file.

So some of my 4-man classes are quite a bit smaller, but most of the
5-man classes are about the same size, since you get some pretty big
values.

I bring up this point in order to discount the differences, effectively
they are the same format.

It has been my experience that on a modern machine with reasonable
memory and a hash table that isn't crowding the machine, that you can do
4-man probes and still go pretty fast.  The relevent tables end up in
the disk cache, and they are 3-10 megabytes, so you can fit a few of
them in memory at once.

If you try to access the 5-man tables at the tips and in the tree, you
will go dog slow.  Mine would be hitting 300-500K nps without those
tables (on my 533 mhz Alpha), but you start adding in tables you can get
under 20K, no problem.

There is a *big* difference between 4-man and 5-man tip probes.

And I have a nice 9 gb SCSI disk, too, so it's not the disk.

It is an open question whether the increased evaluation accuracy (and
ability to accurately prune out some sub-trees) outweighs the bogging
you get when you start to get close to a 5-man ending.  I've never tried
to figure it out for real, I'm just trying to keep my eyes open during
games.

I still use all the tables I can fit on that disk (I have 18 gb of them,
so they don't all fit), but I think it hurts in some cases, like queen
and a few pawns versus queen and a few pawns.  This one especially seems
to go slow, and I've drawn or lost some cases that I think should have
been shifted a column in my favor.

bruce




This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.