-----Inline Message Follows-----
Hi there
Now there's another problem with Rybka 3. It's running allright as long
as I limit the hash size to max 32 MB. If the hash is more then 32 MB
Rybka 3 is down to 1 kn/s and my cpu usage is only a few % and the HD is
running like crazy. All other chess engines work fine even with 1GB hash.
Do you know what's going on?
Regards, Michael
-----------------------------------------
Michael,
Welcome to my world! Hash table usage is very different in Rybka 3 and there
now is a BIG difference between running Rybka 3 under wine on Linux and running
it under Windows. No one can tell me what the problem is, but it is VERY
noticeable. Some do not report the problem at all. I have tried running 32-bit
Rybka 3 under wine using SCIDon 32-bit Ubuntu-8.04, 32-bit Rybka 3 under
Aquarium under wine on 32-bit Ubuntu-8.04, and I've tried running those under
64-bit Ubuntu-8.04. The problem still exists. I have managed to bump hash
tables to 128 MB on my 3 GB RAM machines, but no higher or the problem returns.
I have learned this: when you set hash size to 32 MB, Rybka 3 allocates RAM as
follows: 75MB * number of cores + number of cores * hash-setting. This is
counter-intuitive but it is the way Rybka 3 works. Why that does not cause the
same problem when running under Windows I do not know. Perhaps it is a bug in
wine? I'm running the latest wine too (1.1.2 if memory serves) straight from
winehq.
In the end, I suppose the 32 MB restriction on hash for a 1 GB RAM system
probably makes Rybka 3 run about 30-40 ELO weaker (a guess on my part but I
recall someone saying that each doubling of RAM for hash tables resulted in 8
ELO gain and that would mean that 32 * 4 = 512 (the max MB you could reasonable
expect to use for hash on a 1 GB RAM system).
If anyone figures out a workaround to this, please post to this group!
Roy
----- Original Message ----
From: "[EMAIL PROTECTED]" <[EMAIL PROTECTED]>
To: scid-users@lists.sourceforge.net
Sent: Tuesday, August 19, 2008 3:43:09 PM
Subject: Scid-users Digest, Vol 25, Issue 18
Send Scid-users mailing list submissions to
scid-users@lists.sourceforge.net
To subscribe or unsubscribe via the World Wide Web, visit
https://lists.sourceforge.net/lists/listinfo/scid-users
or, via email, send a message with subject or body 'help' to
[EMAIL PROTECTED]
You can reach the person managing the list at
[EMAIL PROTECTED]
When replying, please edit your Subject line so it is more specific
than "Re: Contents of Scid-users digest..."
Today's Topics:
1. Problem with Rybka 3 an hash size > 32 MB (Michael H?ppenstein)
2. Re: Scid and Rybka 3 (Michal Rudolf)
3. Re: Mask on Tree (Pascal Georges)
4. Re: Mask on Tree (Pascal Georges)
5. Re: Scid and Rybka 3 (Pascal Georges)
-----Inline Attachment Follows-----
2008/8/19 Alexander Wagner <[EMAIL PROTECTED]>
Pascal Georges wrote:
Hi!
I reworked Mask view
Looks very promising promising, indeed! :)
Hope you appreciated the speed up : 10000 % ! Note that Mask info menu gives
stats on hashes that will be removed later.
Just stumbled accross () in commentary. This throws an
exception.
Strange, I tried by importing a game with a fake "()" and added manually "trial
[] ..() .. {} .." without problem (even if I thought [] would lead to problems,
but it works). So could you send me a file or a procedure to reproduce it ?
So far I imported 5000 positions in a Mask file with vars, comments,
evaluations without any problem.
Additionally, could it be that there're some \n written to
the stm file that are not intended.
No, I don't think so. "\n" are appended by the user when entering multi-line
comments (you noticed entry box has changed). On the Tree window only the first
line is displayed, but associated tooltip displays the complete comment.
Adding a marker should probably add the line to the mask
automatically. (Makes sense to me, but maybe I miss an
occasion where it shouldn't?)
For safety I preferred to use a 2 phases process : first add move to Mask, then
add markers /comments to it. This avoids the case where you want to flag in
green e4 but you miss the line and add by error the move just below (but anyway
this kind of error may always happen).
To play arround I would enter my current sor-file. But just
that I do not do it twice: how is it inteded to use
concerning white/black openings?
This is the same logic for maintaining distinct white and black repertoires :
when you review your openings, you do it for one side, and certainly for some
specific lines. So maybe it is clearer for the user to separate things.
Moreover things are not symmetric : you may like playing the French defense
with Black but strictly avoid it with White. Hence the 2 repertoires.
Do you think they should
both go into one file, or should I rather split them? If the
latter is probably cummulative loading of mask files
intended so I could first load the white lines and then the
black ones and have both in the tree?
It is really easy to merge Masks, but current code does not allow it (data is
reseted when opening a Mask). I would prefer to avoid it, to keep things simple
and not confuse users.
Now the main question : are there markers, flags or anything that should be
added ? So far it seems there are enough info, but maybe it would also be
interesting to display score in various colors. For example :
if Average Elo is > 2200 and number of games > 20 and Score > 55%, put it in
green
if Average Elo is > 2200 and number of games > 20 and Score < 45%, put it in red
Pascal
-----Inline Attachment Follows-----
2008/8/19 Alexander Wagner <[EMAIL PROTECTED]>
Pascal Georges wrote:
Hi!
I reworked Mask view
Looks very promising promising, indeed! :)
Hope you appreciated the speed up : 10000 % ! Note that Mask info menu gives
stats on hashes that will be removed later.
Just stumbled accross () in commentary. This throws an
exception.
Strange, I tried by importing a game with a fake "()" and added manually "trial
[] ..() .. {} .." without problem (even if I thought [] would lead to problems,
but it works). So could you send me a file or a procedure to reproduce it ?
So far I imported 5000 positions in a Mask file with vars, comments,
evaluations without any problem.
Additionally, could it be that there're some \n written to
the stm file that are not intended.
No, I don't think so. "\n" are appended by the user when entering multi-line
comments (you noticed entry box has changed). On the Tree window only the first
line is displayed, but associated tooltip displays the complete comment.
Adding a marker should probably add the line to the mask
automatically. (Makes sense to me, but maybe I miss an
occasion where it shouldn't?)
For safety I preferred to use a 2 phases process : first add move to Mask, then
add markers /comments to it. This avoids the case where you want to flag in
green e4 but you miss the line and add by error the move just below (but anyway
this kind of error may always happen).
To play arround I would enter my current sor-file. But just
that I do not do it twice: how is it inteded to use
concerning white/black openings?
This is the same logic for maintaining distinct white and black repertoires :
when you review your openings, you do it for one side, and certainly for some
specific lines. So maybe it is clearer for the user to separate things.
Moreover things are not symmetric : you may like playing the French defense
with Black but strictly avoid it with White. Hence the 2 repertoires.
Do you think they should
both go into one file, or should I rather split them? If the
latter is probably cummulative loading of mask files
intended so I could first load the white lines and then the
black ones and have both in the tree?
It is really easy to merge Masks, but current code does not allow it (data is
reseted when opening a Mask). I would prefer to avoid it, to keep things simple
and not confuse users.
Now the main question : are there markers, flags or anything that should be
added ? So far it seems there are enough info, but maybe it would also be
interesting to display score in various colors. For example :
if Average Elo is > 2200 and number of games > 20 and Score > 55%, put it in
green
if Average Elo is > 2200 and number of games > 20 and Score < 45%, put it in red
(this latest is not related to Masks, but given that I changed the Tree window
so it can display colored text, we may take advantage of it).
Pascal
-----Inline Attachment Follows-----
2008/8/19 Alexander Wagner <[EMAIL PROTECTED]>
Michal Rudolf wrote:
Hi!
> Even better will be guessing proper mode (UCI versus
> *board) of course, but this requires a bit more work.
Some engines are UCI and xboard compatible. User has the choice.
Pascal
_______________________________________________
Scid-users mailing list
Scid-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/scid-users
-------------------------------------------------------------------------
This SF.Net email is sponsored by the Moblin Your Move Developer's challenge
Build the coolest Linux based applications with Moblin SDK & win great prizes
Grand prize is a trip for two to an Open Source event anywhere in the world
http://moblin-contest.org/redirect.php?banner_id=100&url=/
_______________________________________________
Scid-users mailing list
Scid-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/scid-users
