Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Could someone explain this?

Author: Terje Vagle

Date: 09:54:12 08/16/01

Go up one level in this thread


On August 16, 2001 at 12:17:40, Gordon Rattray wrote:

>On August 16, 2001 at 11:50:53, Uri Blass wrote:
>
>>On August 16, 2001 at 07:35:16, Gordon Rattray wrote:
>>
>>>On August 16, 2001 at 05:54:50, Terje Vagle wrote:
>>>
>>>>On August 16, 2001 at 05:21:51, Bertil Eklund wrote:
>>>>
>>>>>On August 16, 2001 at 03:16:59, Terje Vagle wrote:
>>>>>
>>>>>>Hi all,
>>>>>>
>>>>>>When I run Fritz 6 or Junior 6 in infinite mode over night, the NPS drop from
>>>>>>approx. 600KNS from when I start it to approx 75KNS when it has been running for
>>>>>>about 8-10 hours. The clock also starts to blink.
>>>>>>
>>>>>>BUT, when I run Fritz 5.32 in the same conditions, it continues the analysis
>>>>>>with the same KN/s even after several days!
>>>>>>
>>>>>>WHY?
>>>>>>
>>>>>>Thanks in advance.
>>>>>>
>>>>>>Regards
>>>>>>
>>>>>>Terje Vagle
>>>>>
>>>>>Hi!
>>>>>
>>>>>Check the HD activity! It is some kind of overflow of the memory.
>>>>>Try to lower the hash-tables a bit.
>>>>>
>>>>>Bertil
>>>>
>>>>
>>>>Hi,
>>>>
>>>>No Bertil, that is not it I think. I have 512 MB ram on my PC and hashsize of
>>>>128 MB.
>>>
>>>
>>>I have seen the same problems.  Like yourself, I am sure that there is no
>>>endgame tablebases or other disk access coming into effect.  My hashtable size
>>>was 256Mb on a 512Mb RAM machine and it was no where near to being an endgame
>>>position.  I had just put it down to some sort of bug or weakness or the
>>>program?!
>>
>>My guess is that the number of nodes is simply not correct.
>>It is probably no bug and no weakness because the program is not supposed to
>>know to count nodes but to play chess and if the program use 32 bit number for
>>the number of nodes the number is not correct at long time control and there is
>>no problem(at least there is no problem if the program do not use the number of
>>nodes to decide which lines to search first).
>
>I disagree.
>
>If a chess program displays the number of nodes, then it should be correct.
>Otherwise, what is the point in displaying it at all!!  I agree that this isn't
>as significant a bug as, for example, if the program didn't play chess according
>to the rules, but it's still a bug.  I accept that not all bugs have equal
>significance.
>
>If it's a 32 bit number that is wrapping, this is a problem.  The Fritx GUI is
>designed to show node counts.  Fritz is designed to be able to analyse over long
>periods of time.  If this doesn't work, it's a bug.
>
>Gordon

I totally agree with you, Gordon.

Fritz 5.32 can run for days and show total Nodes of analysis of more than 50
billion!

I haven't seen these numbers on Fritz 6 or Junior 6 when they have been
analysing in about the same amount of time....

Terje



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.