New chess engine released: Bringer 1.7

[Gerrit Reubold]

Geschrieben von: / Posted by: Gerrit Reubold at 25 June 2000 02:14:10:
Hi all,
I just copied Bringer 1.7 to my homepage www.reubold.onlinehome.de
The engine was modified in many ways, hopefully it is stronger than previous versions. The GUI supports now the "k-best-mode": display more than one variation in analysis mode. The English help files are still missing, sorry.
Greetings,
Gerrit

[Graham Banks]

Geschrieben von: / Posted by: Graham Banks at 25 June 2000 09:12:32:
Als Antwort auf: / As an answer to: New chess engine released: Bringer 1.7 geschrieben von: / posted by: Gerrit Reubold at 25 June 2000 02:14:10:

Hi all,
I just copied Bringer 1.7 to my homepage www.reubold.onlinehome.de
The engine was modified in many ways, hopefully it is stronger than previous versions. The GUI supports now the "k-best-mode": display more than one variation in analysis mode. The English help files are still missing, sorry.
Greetings,
Gerrit

Hello Gerrit.
I've just downloaded your program so that I can test it in computer-computer matches through WinBoard. I allocate programs 16mb hash where able. For Bringer should I multiply your default settings by 7 to give 8mb, 8mb and 1.6mb to get the closest sensible allocation to 16mb or do you have a better suggestion?
Graham.

[Gerrit Reubold]

Geschrieben von: / Posted by: Gerrit Reubold at 25 June 2000 12:04:09:
Als Antwort auf: / As an answer to: Re: New chess engine released: Bringer 1.7 geschrieben von: / posted by: Graham Banks at 25 June 2000 09:12:32:

Hi all,
I just copied Bringer 1.7 to my homepage www.reubold.onlinehome.de
The engine was modified in many ways, hopefully it is stronger than previous versions. The GUI supports now the "k-best-mode": display more than one variation in analysis mode. The English help files are still missing, sorry.
Greetings,
Gerrit

Hello Gerrit.
I've just downloaded your program so that I can test it in computer-computer matches through WinBoard. I allocate programs 16mb hash where able. For Bringer should I multiply your default settings by 7 to give 8mb, 8mb and 1.6mb to get the closest sensible allocation to 16mb or do you have a better suggestion?
Graham.

Hi Graham,
I recommend setting the hashtables to 10Mb/2Mb/2Mb and, if you have Nalimmov's endgame tablesbases, setup the remaining 2 MB RAM as cache for them (options - settings).
Greetings,
Gerrit

[Graham Banks]

Geschrieben von: / Posted by: Graham Banks at 25 June 2000 13:25:12:
Als Antwort auf: / As an answer to: Hash size for Bringer geschrieben von: / posted by: Gerrit Reubold at 25 June 2000 12:04:09:

Hi all,
I just copied Bringer 1.7 to my homepage www.reubold.onlinehome.de
The engine was modified in many ways, hopefully it is stronger than previous versions. The GUI supports now the "k-best-mode": display more than one variation in analysis mode. The English help files are still missing, sorry.
Greetings,
Gerrit

Hello Gerrit.
I've just downloaded your program so that I can test it in computer-computer matches through WinBoard. I allocate programs 16mb hash where able. For Bringer should I multiply your default settings by 7 to give 8mb, 8mb and 1.6mb to get the closest sensible allocation to 16mb or do you have a better suggestion?
Graham.

Hi Graham,
I recommend setting the hashtables to 10Mb/2Mb/2Mb and, if you have Nalimmov's endgame tablesbases, setup the remaining 2 MB RAM as cache for them (options - settings).
Greetings,
Gerrit

Sorry to trouble you again Gerrit.
I am not using the Nalimov tablebases, so should I allocate 10-3-3, 12-2-2, or 10-4-2? I notice when using 12-2-2 that the evaluation fills up reasonably quickly, so maybe the 10-3-3 or 10-4-2 might be best. I'd appreciate your expert opinion as the programmer.
Regards, Graham.

[Mogens Larsen]

Geschrieben von: / Posted by: Mogens Larsen at 25 June 2000 13:38:33:
Als Antwort auf: / As an answer to: Re: Hash size for Bringer geschrieben von: / posted by: Graham Banks at 25 June 2000 13:25:12:

Sorry to trouble you again Gerrit.
I am not using the Nalimov tablebases, so should I allocate 10-3-3, 12-2-2, or 10-4-2? I notice when using 12-2-2 that the evaluation fills up reasonably quickly, so maybe the 10-3-3 or 10-4-2 might be best. I'd appreciate your expert opinion as the programmer.
Regards, Graham.

My 16Mb configuration used to be 10-4-2 and it seemed to work okay, but it would be interesting to hear what Gerrit thinks. What about 32Mb configuration then? I use 20-8-4 and 2Mb tb cache. Is that the best configuration?
Best wishes...
Mogens

[Graham Banks]

Geschrieben von: / Posted by: Graham Banks at 25 June 2000 13:44:35:
Als Antwort auf: / As an answer to: Hash size for Bringer geschrieben von: / posted by: Gerrit Reubold at 25 June 2000 12:04:09:

Hi all,
I just copied Bringer 1.7 to my homepage www.reubold.onlinehome.de
The engine was modified in many ways, hopefully it is stronger than previous versions. The GUI supports now the "k-best-mode": display more than one variation in analysis mode. The English help files are still missing, sorry.
Greetings,
Gerrit

Hello Gerrit.
I've just downloaded your program so that I can test it in computer-computer matches through WinBoard. I allocate programs 16mb hash where able. For Bringer should I multiply your default settings by 7 to give 8mb, 8mb and 1.6mb to get the closest sensible allocation to 16mb or do you have a better suggestion?
Graham.

Hi Graham,
I recommend setting the hashtables to 10Mb/2Mb/2Mb and, if you have Nalimmov's endgame tablesbases, setup the remaining 2 MB RAM as cache for them (options - settings).
Greetings,
Gerrit

After further experimentation I wonder whether a 6-6-4 0r 7-7-2 would be better as I'm finding that the Pos. hash is not filling quickly at 10-4-2 whereas the Eval. hash is. I'm doing my computer-computer testing at 60 moves in 60 minutes and I'd have to say that in my opinion (although I'm not an expert!) the 6-6-4 allocation seems to do nicely. Not sure how this would go at longer time controls though.
Graham.

[pete]

Geschrieben von: / Posted by: pete at 25 June 2000 14:16:10:
Als Antwort auf: / As an answer to: Re: Hash size for Bringer (further thoughts) geschrieben von: / posted by: Graham Banks at 25 June 2000 13:44:35:

Hi Graham,
I recommend setting the hashtables to 10Mb/2Mb/2Mb and, if you have Nalimmov's endgame tablesbases, setup the remaining 2 MB RAM as cache for them (options - settings).
Greetings,
Gerrit

After further experimentation I wonder whether a 6-6-4 0r 7-7-2 would be better as I'm finding that the Pos. hash is not filling quickly at 10-4-2 whereas the Eval. hash is. I'm doing my computer-computer testing at 60 moves in 60 minutes and I'd have to say that in my opinion (although I'm not an expert!) the 6-6-4 allocation seems to do nicely. Not sure how this would go at longer time controls though.
Graham.

I don't think such a symmetrical setup does make sense .
My experiences with Bringer hash and cache settings :
a.) Pawn tables ( in KB )
2000 seems to be enough whatever you set for the rest .
b.) Position tables ( in KB ) ; Evaluation tables( in KB )
Position tables should be set up as big as possible ; ratio between Position and Evaluation should be about 1:3 or bigger ; it seems to me that much more than 4000 for Evaluation tables is not useful anyway.
c.) Cache ( in kb )
This is only needed if you use the Nalimov tablebases . I did a few experiments with this one . Though there are a few positions where it seemed to help Bringer to have a bigger cache in general I have come to the conclusion that 2000 is enough and more doesn't help at all . Due to the way Bringer uses TBs it is slowed down much when using them , bigger cache doesn't help that much .
d.) When using Bringer under Winboard you should set "Operator time" to 2 sec. as else Bringer has difficulties in fast blitz games without increments ( didn't have a look at the new release yet , might be default now anyway ) .
When playing with Bringer on ICS I personally use
Position tables 12000
Evaluation tables 4000
Pawn tables 2000
Cache 16384 ( this simply because I never changed it )
This seems to work well even at standard time control , Bringer doesn't seem to fill Hash tables extremely fast anyway .
Only a user's opinion , maybe Gerrit will give a different recommendation .
Regards.
pete

[Gerrit Reubold]

Geschrieben von: / Posted by: Gerrit Reubold at 26 June 2000 00:36:17:
Als Antwort auf: / As an answer to: Re: Hash size for Bringer (further thoughts) geschrieben von: / posted by: pete at 25 June 2000 14:16:10:

Hi Graham,
I recommend setting the hashtables to 10Mb/2Mb/2Mb and, if you have Nalimmov's endgame tablesbases, setup the remaining 2 MB RAM as cache for them (options - settings).
Greetings,
Gerrit

After further experimentation I wonder whether a 6-6-4 0r 7-7-2 would be better as I'm finding that the Pos. hash is not filling quickly at 10-4-2 whereas the Eval. hash is. I'm doing my computer-computer testing at 60 moves in 60 minutes and I'd have to say that in my opinion (although I'm not an expert!) the 6-6-4 allocation seems to do nicely. Not sure how this would go at longer time controls though.
Graham.

I don't think such a symmetrical setup does make sense .
My experiences with Bringer hash and cache settings :
a.) Pawn tables ( in KB )
2000 seems to be enough whatever you set for the rest .
b.) Position tables ( in KB ) ; Evaluation tables( in KB )
Position tables should be set up as big as possible ; ratio between Position and Evaluation should be about 1:3 or bigger ; it seems to me that much more than 4000 for Evaluation tables is not useful anyway.
c.) Cache ( in kb )
This is only needed if you use the Nalimov tablebases . I did a few experiments with this one . Though there are a few positions where it seemed to help Bringer to have a bigger cache in general I have come to the conclusion that 2000 is enough and more doesn't help at all . Due to the way Bringer uses TBs it is slowed down much when using them , bigger cache doesn't help that much .
d.) When using Bringer under Winboard you should set "Operator time" to 2 sec. as else Bringer has difficulties in fast blitz games without increments ( didn't have a look at the new release yet , might be default now anyway ) .
When playing with Bringer on ICS I personally use
Position tables 12000
Evaluation tables 4000
Pawn tables 2000
Cache 16384 ( this simply because I never changed it )
This seems to work well even at standard time control , Bringer doesn't seem to fill Hash tables extremely fast anyway .
Only a user's opinion , maybe Gerrit will give a different recommendation .
Regards.
pete

I agree with pete. The position tables are the most important ones for Bringer, 2 - 4 MB for the Eval- and pawn-tables are enough. It doesn't hurt Bringer if the eval and pawn tables are filled soon. I did some test and found out that Bringer needs a certain minimum size of position tables, 1-2 MB will do, if you have more RAM, each doubling of the tables will speed Bringer up by about 10%, not a huge gain for using (e.g.) 32MB instead of 16MB.
If you have the endgame tablebases, I recommend 2-4 MB tablebase cache size, 16 MB seems plenty. Maybe I should make Bringer more RAM-hungry ?
Greetings,
Gerrit