Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: TB cache size recommendations, please!

Author: Jeremiah Penery

Date: 01:39:59 04/13/00

Go up one level in this thread


On April 12, 2000 at 06:05:12, Stefan Meyer-Kahlen wrote:

>On April 11, 2000 at 03:28:03, Jeremiah Penery wrote:
>
>>On April 10, 2000 at 20:28:50, Robert Hyatt wrote:
>>
>>>On April 10, 2000 at 10:00:40, Stefan Meyer-Kahlen wrote:
>>>
>>>>On April 10, 2000 at 09:00:39, Robert Hyatt wrote:
>>>>
>>>>>On April 10, 2000 at 06:10:22, Jouni Uski wrote:
>>>>>
>>>>>>Has anybody found good values for TB cache with 3+4 piece tablebases and 5 piece
>>>>>>set (compressed Nalimov of course)? Is it different? Or is 4 MB always enough?
>>>>>>Is same value good under Fritz and Winboard? thanks
>>>>>>
>>>>>>Jouni
>>>>>
>>>>>
>>>>>Here is what I do on a machine with decent memory, assuming I am playing in
>>>>>a tournament that is playing long time controls:
>>>>>
>>>>>memory=512mb
>>>>>
>>>>>hash=384M
>>>>>hashp=32M
>>>>>cache=32M
>>>>>
>>>>>which leaves 64M for the O/S and filecache buffers.
>>>>>
>>>>>works well for me.  In general, egtb cache is only important in egtb positions,
>>>>>but then it is very important for performance.
>>>>
>>>>I haven't seen much difference for Shredder if I set the tb cache to 1 or 32 or
>>>>256 MB. Any comments?
>>
>>Try analyzing a position with, say, KRBKRN and try different settings for cache.
>> With 1MB, you'll get tons of probes, and contine to get tons of probes.  With
>>256MB, you'll get tons of probes, and then you'll start to get a bunch less
>>pretty soon. (At least this is what should happen...)
>>
>>>Depends on the position...  I ran several tests last year when Eugene first got
>>>this stuff going, and I decided that for ICC level time controls, 32mb was a
>>>good number.  I rarely see my NPS drop drastically.  It all depends on the
>>>position and how many different databases are getting probed...
>>
>>As well as the speed of your HD.
>>
>>I've started using 48MB for cache, and it speeds up dramatically over lower
>>amounts, even though I still get tons of probes sometimes.  If I'm analyzing a
>>position with about 7-8 pieces, with lots of possible transitions into 5-piece
>>endings, I might even put the cache higher, because (on my machine) it speeds up
>>the search a ton.
>
>Did you use Shredder or Crafty for your tests?

I was using Crafty, and I assumed that they work similarly enough that you'd see
similar things happen.  Does Shredder not do a bunch of TB probes when you're
near a TB position?  If you cache these, you'll not have to do the same probe
twice, which saves you a ton on disk I/O (which is what really makes too many TB
probes prohibitively slow in some cases)

It's possible that for Shredder you won't get much of a benefit, because it
probes TBs a lot less or something.  I don't know. :-/

Jeremiah



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.