Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Crafty on dual core

Author: Gregory Owett

Date: 23:25:08 11/07/05

Go up one level in this thread


On November 07, 2005 at 16:07:29, Robert Hyatt wrote:

>On November 07, 2005 at 03:29:49, Gregory Owett wrote:
>
>>On November 06, 2005 at 19:57:38, Robert Hyatt wrote:
>>
>>>On November 06, 2005 at 09:40:13, Bryan Hofmann wrote:
>>>
>>>>On November 06, 2005 at 09:03:04, Andreas Guettinger wrote:
>>>>
>>>>>On November 06, 2005 at 07:48:20, Bryan Hofmann wrote:
>>>>>
>>>>>>On November 06, 2005 at 07:29:56, Gregory Owett wrote:
>>>>>>
>>>>>>>On November 06, 2005 at 07:13:26, Bryan Hofmann wrote:
>>>>>>>
>>>>>>>>On November 05, 2005 at 08:47:34, Gregory Owett wrote:
>>>>>>>>
>>>>>>>>>When Crafty plays vs another engine, in adjustment "ponder=off", it is pondering
>>>>>>>>>and the taskmrg indicates 99% to Crafty and 0% to the other. Even if its
>>>>>>>>>adversary plays, the indicator gives him 50% and 50% to the other engine,
>>>>>>>>>whereas Crafty should have 0%.  Does it profit from the time of reflection of
>>>>>>>>>the other engine?  If not, why it does not release the processor?
>>>>>>>>>
>>>>>>>>>Gregory
>>>>>>>>
>>>>>>>>If you are using a version prior to 19.16 then you will see this behavior. In
>>>>>>>>version prior to 19.16 Crafty does not release the parallel CPUs, even tho it is
>>>>>>>>not using them (pondering/thinking).
>>>>>>>>
>>>>>>>>
>>>>>>>>Bryan
>>>>>>>
>>>>>>>That explains all. I used the version 19.11 SE.
>>>>>>>It is one of the best, isn't it?  Is there a better according to you?
>>>>>>>
>>>>>>>Thanks.
>>>>>>>
>>>>>>>Gregory
>>>>>>
>>>>>>
>>>>>>I just did some checking, version 19.16 and higher will terminate the additional
>>>>>>threads only when the game has ended. It will still hold on to the additional
>>>>>>threads during the game even though it is not think/pondering the move. So the
>>>>>>issue described still exists in all SMP version of Crafty.
>>>>>>
>>>>>>
>>>>>>Bryan
>>>>>
>>>>>I don't see the problem there. Crafty could hold 100 threads and still use 0% of
>>>>>CPU time. It dosen't need to release the threads when the opponent is thinking.
>>>>>
>>>>>regards
>>>>>Andy
>>>>
>>>>If you have a dual core and wanted to run a match with with SMP engines with
>>>>pondering off it would be a issue. Crafty even with ponder off will keep CPUs 1
>>>>busy, only CPU 0 would be free for the other engine.
>>>>
>>>>
>>>>Bryan
>>>
>>>
>>>This is correct.  I use a "thread pool" concept where I fire up the extra
>>>threads on the first search, and those processes do not terminate until the game
>>>ends, to avoid the repeated overhead of starting/stopping threads throughout the
>>>game.
>>>
>>>Moral of this is if you have two cpus, use each engine with just one CPU and
>>>this won't be a problem, then you can also play ponder=on since each program can
>>>use 100% of one cpu all the time...
>>
>>If I set ponder=on, there would not be a problem at the end-game, when the two
>>programs call the tablabase at the same time?
>>
>>Gregory
>
>
>Nope.  Windoes caches filesystem reads anyway, so things will actually be faster
>for both programs...

I see, thank you.



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.