Author: Kaj Soderberg
Date: 22:36:12 06/22/05
Go up one level in this thread
On June 22, 2005 at 07:45:22, Eelco de Groot wrote:
>On June 20, 2005 at 15:23:09, Kaj Soderberg wrote:
>
>>On June 18, 2005 at 14:52:44, Eelco de Groot wrote:
>>
>>>On June 18, 2005 at 13:38:03, Kaj Soderberg wrote:
>>>
>>>>On June 18, 2005 at 05:12:03, Ed Schröder wrote:
>>>>
>>>>>On June 18, 2005 at 02:47:08, Kaj Soderberg wrote:
>>>>>
>>>>>>On June 17, 2005 at 16:57:25, Ed Schröder wrote:
>>>>>>
>>>>>>>On June 17, 2005 at 16:44:35, Eelco de Groot wrote:
>>>>>>>
>>>>>>>>On June 17, 2005 at 16:22:22, Ed Schröder wrote:
>>>>>>>>
>>>>>>>>>On June 17, 2005 at 15:52:03, Kaj Soderberg wrote:
>>>>>>>>>
>>>>>>>>>>Thank you Ed,
>>>>>>>>>>
>>>>>>>>>>Now i can use my old favourite good tactical, and yet positionally sound, engine
>>>>>>>>>>and compare it with "Silver".
>>>>>>>>>>
>>>>>>>>>>By the way, i have noticed that Pro Deo sometimes (not often of course)in the
>>>>>>>>>>end sort of crashes in analysis mode. The suggested move seems correct and the
>>>>>>>>>>evaluation too, but in the suggested variation it gives away tons of material
>>>>>>>>>>for no reason at all and therefore should lose. I found this out at 400 mb hash
>>>>>>>>>>and then checked it on 200mb to see if it is maybe a hash problem, but the
>>>>>>>>>>problem still occurs. Rebel 12 did the same (again not often) in analysis mode.
>>>>>>>>>>In DOS i have never seen this occur.
>>>>>>>>>>
>>>>>>>>>>I run the beast under ChessPartner with Intel Centrino at 2ghz end one gig of
>>>>>>>>>>memory. Any idea what could be the case, or is it just a problem with displaying
>>>>>>>>>>the right variation? I seem to remember that Thorsten Czub once run into
>>>>>>>>>>something similar.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>I occasionally face the same (a crash) under ChessPartner after a long time in
>>>>>>>>>analysis mode. Annoying indeed and since it is not reproducible I have no idea
>>>>>>>>>what possibly might be the cause.
>>>>>>>>>
>>>>>>>>>Ed
>>>>>>>>
>>>>>>>>
>>>>>>>>Hallo Ed, I think I can report something very similar:
>>>>>>>>
>>>>>>>>I have not seen any "sort of crashes" but twice ProDeo 1.1 really crashed on my
>>>>>>>>computer in analysis mode. Both times it was just before finishing a 24 hour
>>>>>>>>analysis, so I expect it to be some sort of overflow, of the nodecounters maybe?
>>>>>>>>Windows terminates Pro Deo and asks to file a report. What I can see in
>>>>>>>>Dr.Watson's log that comes with Win XP is that both times there occurs a
>>>>>>>>division by zero. But what exactly gets divided is not clear to me.
>>>>>>>
>>>>>>>Thanks Eelco, the "24 hours" item makes me hear all kind of bells ringing. It's
>>>>>>>perhaps so that after 24 hours the internal clock is set to zero. I must
>>>>>>>check...
>>>>>>>
>>>>>>>Ed
>>>>>
>>>>>>At the speed of my machine, the earliest "crash" i witnessed in the display was
>>>>>>after 15 minutes and 50 seconds. You mention this problem occurs when running in
>>>>>>ChessPartner. Would that suggest that this problem does not occur in other GUIs?
>>>>>
>>>>>I wouldn't know, I exclusively use my DOS version or CP for manual testing.
>>>>>
>>>>>Ed
>>>>
>>>>I checked it under Chessbase. No direct crash, but suddenly there was no
>>>>(new)engine info when reaching a new ply. The new ply was not reported. Maybe it
>>>>the "crashed".
>>>>
>>>>I may downlodad the Shredder interface and check wat the engine does there.
>>>>
>>>>I would not mind if you would give the Pro Deo downloadable as a DOS version
>>>>also. I would use it anyway.
>>>>
>>>>Best regards,
>>>>
>>>>Kaj
>>>
>>>Not speaking for Ed, but I think I've seen that when I tried Pro Deo 1.1 in the
>>>Shredder interface as UCI-engine, but only -I hope!- when Pro Deo could not find
>>>a proper personality file. In param.txt there was no personality loaded. The
>>>program played considerably weaker and I think a couple of times a new ply was
>>>reached -I could see that in Uci-debug mode from Shredder- but not communicated
>>>as a new iteration.
>>>
>>>I hope this was only in the version of Pro Deo that gets loaded if it cannot
>>>find a personality file. It is possible to check that with param.txt
>>>
>>> Regards, Eelco
>>
>>Hello there,
>>
>>The bad news is that Pro Deo 1.1 and Rebel 12 behave the same way under the
>>Shredder interface (beautiful interface though!).
>>
>>The good news is that Pro Deo 1.1 and Rebel 12 are doing just fine and stable
>>under ChessPartner when running with 60mb of hash. Tried it (just because it's
>>the default value) for hours and hours with my new Centrino and my little bit
>>older AMD, and without problems. Maybe the "crash" we are facing is a hash table
>>problem after all?!
>>
>>Best regards,
>>
>>Kaj
>
>Hallo Kaj,
>
>I am glad that Pro Deo 1.1 now is working well on your Centrino and AMD! I am
>not completely sure that using 60 Mb HT completely solves all troubles, if that
>means no real crashes or almost crashes. I have not yet tried that, using no
>more than 60 Mb, but just have a feeling it could be more complex. But not
>enough data myself...
>
>Also I am afraid Ed will not yet have information enough about the bad lines you
>saw, to be able to maybe do something about it? I mean what position it was with
>how much hash and what lines you saw.
>
>I know I should also try to put Pro Deo past the "24 hour barrier" if indeed
>there is such a thing. But with the warm weather over here in Holland I rather
>not let the computer run for such a very long time! Also I'm running some
>matches at the moment that keep the Athlon occupied.
>
>I am pitting three variations of the Ripley settings, with the same Noomen
>openings, against Gandalf 6.01 and with 200 Mb HT for the engines. At the end
>that should give enough material to at least calculate a reliable rating versus
>Gandalf 6.01, and also see if there is an adverse effect of using 200 Mb HT.
>
>The variations in the Ripley/Vulcan {9iXX} settings are not very consequential I
>think, that will make it interesting to see if they have different results or
>games, and for a rating I can also throw the results in one bin to get a smaller
>error margin.
>
>Best regards, sorry for the delay in answer!
> Eelco
Eelco,
Interesting to see what the beast does with Gandalf. I only have Gandalf 5.1,
being such a favourite of mine because it from time to time finds
moves/solutions that really no engine or human for that matter would think of;
sometimes also difficult tactics in a second or so. Until now it seems that G6
is better but a little bit boring in comparison. But we shall see. Maybe i'll go
for G6 in the end anyway.
Going on holiday now and Rebel and Pro Deo stuff will be discussed again towards
the end of July. Then i will try to think of disciplined saving of problem
situations. For myself i just play around with the engine for fun and do not
note stuff. But i will start doing so for a while when i'm back, so i can give
input for Ed.
Good luck with the match.
Best regards,
Kaj
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.