Gothmog's hashsize...

[Roger Brown]

Geschrieben von:/Posted by: Roger Brown at 01 April 2004 04:12:28:

Hello all (especially Tord),
Not that Gothmog needs any more publicity than it already enjoys - think of how many times the author has to post his apologies for crappy code etc. Yeah right.
Anyway, as a WB engine I notice that Gothmog is not taking on the hashsize indicated in gothmog.ini. I have copied my setting for your convenience.

; Use tournament book, "yes" or "no":
tbook yes
; Use own book by default in UCI mode, "yes" or "no":
ownbook-default yes
; Print all scores from white's point of view in xboard mode
white-score no
; Use short algebraic notation in xboard mode:
san yes
; Piece letters in your favorite language:
piece-letters PNBRQK
; Hash table size in MB:
hash 50
; Hash size for use in UCI mode when the GUI tells the engine to use
; only 1MB. This is a workaround for a bug in the Fritz GUI.
fritz-hash 48
; Style of play, "normal", "active", "aggressive" or "suicidal":
style active
; Degree of selectivity, from 0 to 10:
selectivity 6
; Checks in qsearch, "yes" or "no":
qsearch-checks yes
; Use recursive nullmove pruning, "yes" or "no":
nullmove yes
; Nullmove reduction factor, should be 1, 2 or 3:
R 3
; Evaluation grain size, should be 1, 2, 4 or 8:
grain-size 4
; Resign, "yes" or "no":
resign yes
; Location of Nalimov tablebases:
tbpath C:\TB\
; Use internal iterative deepening, "yes" or "no".
iid yes


Taskinfo is indicating a hashsize of approximately 9 Mb.....
I have seen this before with another WB/UCI engine. I would like to know how to correct it as it hardly seems fair to have Gothmog compete on such uneven playing field. Any suggestions would be welcome. A similar thing ocurs regardless of whether the engine is used with the Winboard or UCI engine protocols.
Can someone offer an explanation for this?
Later.

[Olivier Deville]

Geschrieben von:/Posted by: Olivier Deville at 01 April 2004 08:33:25:
Als Antwort auf:/In reply to: Gothmog's hashsize... geschrieben von:/posted by: Roger Brown at 01 April 2004 04:12:28:

Hello all (especially Tord),
Not that Gothmog needs any more publicity than it already enjoys - think of how many times the author has to post his apologies for crappy code etc. Yeah right.
Anyway, as a WB engine I notice that Gothmog is not taking on the hashsize indicated in gothmog.ini. I have copied my setting for your convenience.

; Use tournament book, "yes" or "no":
tbook yes
; Use own book by default in UCI mode, "yes" or "no":
ownbook-default yes
; Print all scores from white's point of view in xboard mode
white-score no
; Use short algebraic notation in xboard mode:
san yes
; Piece letters in your favorite language:
piece-letters PNBRQK
; Hash table size in MB:
hash 50
; Hash size for use in UCI mode when the GUI tells the engine to use
; only 1MB. This is a workaround for a bug in the Fritz GUI.
fritz-hash 48
; Style of play, "normal", "active", "aggressive" or "suicidal":
style active
; Degree of selectivity, from 0 to 10:
selectivity 6
; Checks in qsearch, "yes" or "no":
qsearch-checks yes
; Use recursive nullmove pruning, "yes" or "no":
nullmove yes
; Nullmove reduction factor, should be 1, 2 or 3:
R 3
; Evaluation grain size, should be 1, 2, 4 or 8:
grain-size 4
; Resign, "yes" or "no":
resign yes
; Location of Nalimov tablebases:
tbpath C:\TB\
; Use internal iterative deepening, "yes" or "no".
iid yes


Taskinfo is indicating a hashsize of approximately 9 Mb.....
I have seen this before with another WB/UCI engine. I would like to know how to correct it as it hardly seems fair to have Gothmog compete on such uneven playing field. Any suggestions would be welcome. A similar thing ocurs regardless of whether the engine is used with the Winboard or UCI engine protocols.
Can someone offer an explanation for this?
Later.

Hi Roger
This problem was already reported. It seems that this is merely a display problem. If you look at the Free physical mem, you'll see that Gothmog is taking more than 9 Mb.
Olivier


ChessWar

[Kurt Utzinger]

Geschrieben von:/Posted by: Kurt Utzinger at 01 April 2004 09:09:53:
Als Antwort auf:/In reply to: Re: Gothmog's hashsize... geschrieben von:/posted by: Olivier Deville at 01 April 2004 08:33:25:

Hello all (especially Tord),
Not that Gothmog needs any more publicity than it already enjoys - think of how many times the author has to post his apologies for crappy code etc. Yeah right.
Anyway, as a WB engine I notice that Gothmog is not taking on the hashsize indicated in gothmog.ini. I have copied my setting for your convenience.

; Use tournament book, "yes" or "no":
tbook yes
; Use own book by default in UCI mode, "yes" or "no":
ownbook-default yes
; Print all scores from white's point of view in xboard mode
white-score no
; Use short algebraic notation in xboard mode:
san yes
; Piece letters in your favorite language:
piece-letters PNBRQK
; Hash table size in MB:
hash 50
; Hash size for use in UCI mode when the GUI tells the engine to use
; only 1MB. This is a workaround for a bug in the Fritz GUI.
fritz-hash 48
; Style of play, "normal", "active", "aggressive" or "suicidal":
style active
; Degree of selectivity, from 0 to 10:
selectivity 6
; Checks in qsearch, "yes" or "no":
qsearch-checks yes
; Use recursive nullmove pruning, "yes" or "no":
nullmove yes
; Nullmove reduction factor, should be 1, 2 or 3:
R 3
; Evaluation grain size, should be 1, 2, 4 or 8:
grain-size 4
; Resign, "yes" or "no":
resign yes
; Location of Nalimov tablebases:
tbpath C:\TB\
; Use internal iterative deepening, "yes" or "no".
iid yes


Taskinfo is indicating a hashsize of approximately 9 Mb.....
I have seen this before with another WB/UCI engine. I would like to know how to correct it as it hardly seems fair to have Gothmog compete on such uneven playing field. Any suggestions would be welcome. A similar thing ocurs regardless of whether the engine is used with the Winboard or UCI engine protocols.
Can someone offer an explanation for this?
Later.

Hi Roger
This problem was already reported. It seems that this is merely a display problem. If you look at the Free physical mem, you'll see that Gothmog is taking more than 9 Mb.
Olivier

There are also other engines like Deep Sjeng where TaskInfo is unable
to display correct use of memory.
Kurt

[Fabien Letouzey]

Geschrieben von:/Posted by: Fabien Letouzey at 01 April 2004 10:45:15:
Als Antwort auf:/In reply to: Re: Gothmog's hashsize... geschrieben von:/posted by: Kurt Utzinger at 01 April 2004 09:09:53:

There are also other engines like Deep Sjeng where TaskInfo is unable
to display correct use of memory.
Kurt

Deep Sjeng uses shared memory (for multi-process purposes), this is a separate issue.
Fabien.

[Tord Romstad]

Geschrieben von:/Posted by: Tord Romstad at 01 April 2004 14:02:55:
Als Antwort auf:/In reply to: Gothmog's hashsize... geschrieben von:/posted by: Roger Brown at 01 April 2004 04:12:28:

Hello all (especially Tord),
Not that Gothmog needs any more publicity than it already enjoys - think of how many times the author has to post his apologies for crappy code etc. Yeah right.
Anyway, as a WB engine I notice that Gothmog is not taking on the hashsize indicated in gothmog.ini.
I have copied my setting for your convenience.

Hello Roger!
I am fairly sure it does, even though taskinfo doesn't report more than
9 MB of memory used. I have hardly any technical knowledge about Windows,
but I think the reason for the apparently low memory consumption is that
Gothmog is not a "real" Windows application, but a Cygwin application.
My guess is that Cygwin apps somehow allocate memory in a different way
than ordinary Windows applications, and that this makes taskinfo confused.
If Gothmog were really using only 9 MB, it would probably play below patzer
level. Because of the way my search works, Gothmog needs big hash tables
in order to work well.
Nothing wrong with the settings, as far as I can see.
Tord