Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Correction, Sorry Re: Is the NPS tend to grow at the end of the game?

Author: Ed Schröder

Date: 11:17:49 07/25/00

Go up one level in this thread


On July 25, 2000 at 12:32:59, Graham Laight wrote:

>On July 25, 2000 at 10:30:01, Ed Schröder wrote:
>
>>On July 25, 2000 at 06:55:22, Graham Laight wrote:
>>
>>>On July 25, 2000 at 01:58:54, Ed Schröder wrote:
>>>
>>>>
>>>>Hash tables and NPS generally have very little in common. Lowering the
>>>>hash table size will increase NPS with a few percent because less hash
>>>>table code is executed.
>>>>
>>>>Ed
>>>
>>>If this is right, then the best thing to do is to have no hash tables whatsoever
>>>- this will maximise your NPS and simplify your code!
>>
>>No hash tables will result in a speed drop of a factor 3-4 average. The
>>hash table code is in the search module and not in the evaluation module.
>>
>>Ed
>
>Sorry, but I'm getting confused here. Let me try to state as simply as possible
>what I think you're saying, and if it's wrong, somebody can correct me:
>
>* hash tables are used to store positions which have already been evaluated
>
>* sometimes, in the search, a position will come up which has already been
>looked at previously
>
>* if so, that branch of the search need not be done again, because the results
>are already in the hash table
>
>* thus, using hash tables, you'll do fewer NPS, but you'll still be better off
>because you won't be wasting time searching branches that have already been
>searched
>
>-g

The increase of NPS with no hash table explains as follows:
- less code to execute for the processor.
- no waste of the PC's data cache. The bigger the hash table the more
worse for the L1 data cache.

Not that it matters we are only talking on a few percent here while the
gain in speed is about 300-400 percent.

Ed



This page took 0.01 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.