Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Max CacheSize on nalimov tables in Shredder

Author: Robert Hyatt

Date: 13:49:54 04/30/04

Go up one level in this thread


On April 30, 2004 at 16:25:29, Eugene Nalimov wrote:

>On April 30, 2004 at 16:24:39, Robert Hyatt wrote:
>
>>On April 30, 2004 at 14:42:29, Vincent Diepeveen wrote:
>>
>>>On April 30, 2004 at 13:44:17, Ingo Bauer wrote:
>>>
>>>>On April 30, 2004 at 13:36:44, Ingo Bauer wrote:
>>>>
>>>>>On April 30, 2004 at 09:53:01, Karl-Johan Olsen wrote:
>>>>>
>>>>>>btw. it's in the classical interface...
>>>>>
>>>>>Hi
>>>>>
>>>>>I never checked this and can't at the moment (will later) but it is completly
>>>>>useless. 64 is more than enough, let the OS do the caching job!
>>>>>
>>>>>Bye
>>>>>Ingo
>>>>
>>>>OK, I could not wait:
>>>>
>>>>CPU0: AuthenticAMD x86 Family 6 Model 10 Stepping 0 2505 MHz
>>>>GUI: Tablebases with 5 pieces found! [Cache: 999 MB + internal 13.67 MB]
>>>>Engine: Shredder 8 (64 MB)
>>>>by Stefan Meyer-Kahlen
>>>>  4/04	 0:00 	  +M11	1.Ne6 Rg6 (95) 5
>>>>best move: Nc7-e6 time: 0:00.016 min  n/s: 7.687  CPU 93.7%   n/s(1CPU): 8.203
>>>>nodes: 123 TB: 174
>>>>
>>>>There seems to be a limitation of 999 MB for the TBs Cache, and the 5 Pcs at
>>>>least are working fine. But again 999 is completly useless! The OS is doing the
>>>>caching job allready. Use a bit more for Engine-Hash and let 100-200 MB Memory
>>>>free for the OS. Done!
>>>>
>>>>Ingo
>>>>
>>>>Ingo
>>>
>>>The OS might cache it Ingo, but won't decompress it for you which also takes
>>>quite some time. A bigger EGTB cache works better than letting the OS find the
>>>golden coins.
>>>
>>>Best regards,
>>>Vincent
>>
>>Current egtb cache caches compressed blocks.  So this is a moot point.
>
>Unfortunately no. That is in my "TODO" list.
>
>Thanks,
>Eugene

So you are caching decompressed stuff?  IE a buffer to read into, then
decompress into the cache???  However, the original point is still valid, that
this is not a big issue.   Decompression compared to I/O is tiny.

>
>>Decompression is _not_ the bottleneck.  From actual testing rather than
>>guessing...



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.