Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Help Eugene Nalimov

Author: Robert Hyatt

Date: 15:05:36 03/26/03

Go up one level in this thread


On March 26, 2003 at 11:10:24, Uri Blass wrote:

>On March 26, 2003 at 10:55:19, Robert Hyatt wrote:
>
>>On March 26, 2003 at 08:20:04, Patrick Götz wrote:
>>
>>>Hello
>>>
>>>is there any way to help Eugene Nalimov with my AMD XP 2000+ 1 GB RAM ?
>>>Perhaps we can see the 6 man TBs with additional Computerpower a little bit
>>>sooner .
>>>
>>>regards
>>>Patrick
>>
>>
>>You don't have enough compute power, enough main memory, and almost certainly
>>not
>>enough disk space either.  :)
>
>I am not sure.
>The main question is how much main memory is needed for generating
>one tablebase with 5 pieces and one pawn(I assume that not all of them were
>generated)
>


I don't think those are very interesting.  The code won't generate them if I
recall
Eugene's comments correctly.  I therefore assumed that this was help with the 6
piece files and those require a _horse_ of a machine.



>For example take the KBB vs KNP tablebases.
>Assume for the discussion it was not generated(in case that it was generated it
>is possile to take another tablebases with one pawn).
>
>We have the following questions:
>1)How much disk space is needed to store all the sons of this table?
>2)How much main memory is needed to generate only the case of pawn in the 7th
>(the case of pawn in the 6th can be generated later)?
>3)How much time is needed to do it on XP 2000+ 1 GB?
>
>Uri

All I can say without thinking about it carefully is that the answer to all
questions is
"big".

The tables have to be uncompressed.  That turns the 6 piece files into _huge_
things.  IE
64^6 bytes roughly, not counting the symmetry and built-in savings for Eugene's
indexing
scheme.  But if that is a factor of (say) 16, then the size is _still_ huge, as
2^36 / 2^4 = 2^32
which is 4 gigabytes, a problem in its own right.

The temp files are going to be in that range as well, and the more that fits
into memory,
the better.  Also, I believe that Eugene mmaps() the existing tables into the
virtual address
space meaning that a 64 bit address space may be needed, leaving PCs out in the
cold
completely.

He might respond differently however..




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.