Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Deep Fritz & SMP.....

Author: Robert Hyatt

Date: 15:16:15 12/04/01

Go up one level in this thread


On December 04, 2001 at 14:40:42, Andrew Dados wrote:

>On December 04, 2001 at 11:56:59, Robert Hyatt wrote:
>
>>On December 04, 2001 at 04:04:37, Jonas Cohonas wrote:
>>
>>>On December 03, 2001 at 22:11:59, K. Burcham wrote:
>>>
>>>>
>>>>
>>>>it seems you got about a 30% speed up in this example, even though it misses the
>>>>move.
>>>>also my task manager shows that deep fritz is using 100% of both processors.
>>>>
>>>>but what really bothers me---if i log into the fritz7 server, even though
>>>>the software is running everything, time etc., it is still using my engine
>>>>in my pc for the game. so i dont understand the big speedup at fritz7 server.
>>>>
>>>>kburcham
>>>
>>>I got a big speedup with DF on the F7 server in the first game played, after
>>>that it was back to "normal". There must be a bottleing up of something in DF
>>>because the speedup almost get's worse for each game.
>>>
>>>It would be nice with an upgrade that would fix tthis. BTW i spoke to a guy at
>>>the F7 server who were testing a beta version of DF7 and he said he got a 35%
>>>speedup.
>>>
>>>To anyone who would be qualified to answer this: could the lack of or poor
>>>speedup be caused by the fact that DF was written so it could be used with 1
>>>processor too?
>>>
>>>Regards
>>>Jonas
>>
>>
>>I don't see how.  IE the SMP version of crafty will use one or two or four
>>cpus with no problems.  And the SMP version is roughly .005% slower than the
>>non-SMP version when run on a single processor.  The only reason I distribute
>>two versions of crafty executables (SMP and non-SMP) is to prevent someone from
>>using the SMP version on a single cpu, and telling it to use two threads.  That
>>will absolutely kill performance.
>
>Can't you just auto-detect this? Most OSes provide cpu number easily.
>-Andrew-


Yes, but users typically don't like it.  First, there is no ANSI compliant way
to determine the number of processors, so it becomes a portability issue.
Second, in many cases users don't want the machine to "see two and use two"
automatically (some programs did this in the past and it caused a lot of
complaints.)

For me the issue is portability above all else...



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.