Rybka3, Arena & Hashtable szie

[chesstango]

Kah Huat, Koh

forget this stuff; i have never found anybody who really wants to share some "secrets", or tips, or settings, nothing.....

[Karmazen & Oliver]

is strange that ??? OK. It´not strage. it´s A BUG..

more RAM always is better for fast and better plys search...

i doubt that...

My thoughts about it:

It makes no sense to use e. g. 512 MB HT on a slow computer, as it has no chance to fill all the memory! It would simply waste too much processing time with filling and managing the Hash Tables.

And if the correct move is found early then small HTs should perform better.

Robert

a slow computer

?? TODAY don´t exits cpu that have problems with 256 MB or 512 MB...

It would simply waste too much processing time with filling and managing the Hash Tables.

This time, ( how many time ?= 1 second ? , less than one second, more of one second, neither it cares, it is worthless for normal rhythms of analysis and fo the game, (alone it is important for game rhythms for all the plays very quick 1 min / game or 3 min / game.), alone it is used once, as a lot in each move-play, if there are really already problems and differences in a single move-play and the ram it has not been reset to calculate the following one move, then it is a bug

if the correct move is found early then small HTs should perform better.

on analisys game ? it´s false. only in super fast game... 1 min / all game.

[Karmazen & Oliver]

Hi Robert,

I am using an Intel Core 2 Duo CPU, 2.20 GHz. I don't think it is a slow cpu....
Best Regards,
Kah Huat, Koh
- from Singapore

of course ¡ That is better than my old pc. ( it´s old but it if enough )

( my pc is a intel D 820 D, underclocking 2,1 Ghz... normal speed 2,8, in overclockin 3,5 Ghz ) ram speed 400-533-600 Mhz... with 2 Gb RAM, and HDD western digital raptor 150 gb X 2.

best regards, from spain. Oliver.

[Karmazen & Oliver]

Kah Huat, Koh

forget this stuff; i have never found anybody who really wants to share some "secrets", or tips, or settings, nothing.....

I understand what you mean

go, it is a very pessimistic comment, what is certain it is that there are people that if they have not been happened to them, then it doesn't serve.

but I have found a lot of interesting and coherent people...

positivism. regards

[chesstango]

Thx a lot Oliver. I am an optimistic man but a real man too...... I agree with your comments, but lets say that there are just a few of them over the chess fora......

[Gerd Lahnstein]

As promised here the results of my futher testing...

Full report (in German):
http://www.schachfeld.de/f202/rybka-3-s ... tung-7874/

In short for non-Germans:

Did another 2 tests, again repeated the tests at least two times, some of the shorter tests up to 10 times, always with IDENTICAL result.

Position #2

FEN: 2b1r1k1/q5bn/2p1pB2/2PpPP2/1p1P3N/r1n2K2/3Q1P1N/1B4RR w - - 0 18
taken from here:
http://www.schachfeld.de/f18/eigenen-partie-2133/

best move: Lxg7

time to find the best move depending on hash size
(sorted by time)

Found within minutes:
256 MB: 13+ 0:06:35 5.672.204 14.723 +1,51 18.Lxg7
32 MB: 14+ 0:08:46 7.716.014 15.012 +1,38 18.Lxg7

Found after hours:
512 MB: 16+ 2:16:19 107.449.958 13.453 +3,04 18.Lxg7
128 MB: 16+ 3:33:14 175.889.949 14.077 +2,89 18.Lxg7
64 MB: 16+ 4:01:33 203.065.652 14.347 +3,01 18.Lxg7

For Analysis details see:
http://www.schachfeld.de/attachment.php ... 1225073868


Position #3

FEN: 1rr1b1k1/1pq1bp2/p2p1np1/4p3/P2BP3/2NB2Q1/1PP3PP/4RR1K w - - 0 1
taken from:
"Der brillante Schachzug (9)", Walter Eigenmann
http://glareanverlag.wordpress.com/cate ... schachzug/

best move: Txf6

time to find the best move depending on hash size
(split by found/not found, LOL)

Move found (repeated 3 times):
32 MB: 18+ 1:07:25 61.359.291 15.530 +0,63 1.Txf6

Move NOT FOUND:
64 MB: 20+ 2:56:16 149.520.839 14.477 +0,61 1.La7
128 MB: 20 2:10:27 115.014.635 15.046 +0,63 1.La7
256 MB: 20+ 2:57:47 151.127.625 14.507 +0,62 1.La7
512 MB: 21+ 5:37:59 351.187.209 15.060 +1,07 1.La7

For Analysis details see:
http://www.schachfeld.de/attachment.php ... 1225073868


Summary:
It becomes more and more obvious that a larger hash not necessarily means faster/better results. With rybka 3 I don't see any reason to increase the hash size, at least on my system.

@ Regarding "slow computer and filling hash...", believe that especially a slower system would benefit much more from a larger hash because a slower system would be much more under load with all these extensive calculations than to manage its hash, however it certainly depends on the calculations and the hash design.

[Ovyron]

Ok, I'm reporting this at Rybka forum.

[Gerd Lahnstein]

Ok, I'm reporting this at Rybka forum.

Many thanks, looks like a good idea.

[Kohflote]

Hi and thank you all!! Hope someone could clarify and resolve this mystery (bug?)

Best Regards,
Kah Huat, Koh
- from Singapore

[Robert Weck]

a slow computer

?? TODAY don´t exits cpu that have problems with 256 MB or 512 MB...

I know many people that use P III with around 600-800 MHz for chess; far more than enough for playing against Rybka and co.!

Of course you can use 512 MB for HT, but in my eyes its only a waste of memory, if you do games with less than 5 mins/move!

Nevertheless i only wanted to point out, that i doubt, that the statement "more is always better" is really true!

If you say: "that is a bug", you must be either an experienced programmer or you tested other engines with various HT sizes; if not, i can't take your statement for serious...

if the correct move is found early then small HTs should perform better.

on analisys game ? it´s false. only in super fast game... 1 min / all game.

Have you ever checked, how long it takes to fill hash tables?

Here i have a notebook with a Pentium M 1,7 GHz and 512 MB; with Toga 3.1.2 it takes around 2 mins to fill just 128 MB.

too much for super fast games...


Robert

P.S.: not so long ago, i read a posting in a forum (i don't remember which forum it was, sorry; maybe the german CSS-forum?) who did various tests and got also better results with 128 than with 256 or 512 MB...