Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Crafty Opteron Performance Data

Author: Robert Hyatt

Date: 13:45:52 06/19/03

Go up one level in this thread


On June 18, 2003 at 18:39:00, Jeremiah Penery wrote:

>On June 18, 2003 at 06:05:52, Aaron Gordon wrote:
>
>>On June 18, 2003 at 00:23:04, Jeremiah Penery wrote:
>>
>>>On June 17, 2003 at 19:01:25, Aaron Gordon wrote:
>>>
>>>>On June 17, 2003 at 01:41:59, Jeremiah Penery wrote:
>>>>
>>>>>On June 16, 2003 at 23:17:12, Jay Urbanski wrote:
>>>>>
>>>>>>On June 15, 2003 at 19:04:02, Vincent Lejeune wrote:
>>>>>>
>>>>>>>Could we get result for 2 crafty bench in same time (one by CPU), please ?
>>>>>>
>>>>>>Sure.
>>>>>>
>>>>>>CPU 0
>>>>>>Raw nodes per second: 1150676
>>>>>
>>>>>>CPU 1
>>>>>>Raw nodes per second: 1150676
>>>>>
>>>>>
>>>>>That's almost impossible to believe.
>>>>
>>>>My dual Athlon shows the same almost.. check out my previous post to Hyatt.
>>>>
>>>>http://www.talkchess.com/forums/1/message.html?300395
>>>>
>>>>1.995x speedup running two copies of Crafty, one per cpu. 1.6x speedup with 1
>>>>Crafty, 2 cpus. This is a Crafty problem however and it's also occuring on
>>>>Hyatt's dual Xeon system.. he's currently looking into the problem.
>>>
>>>Your system shows *identical* NPS for 2 concurrent bench tests?
>>
>>Yes. I think Crafty has an inaccurate timer. Like taking 28718972 nodes (for
>>example) and dividing it by 15, 16, 17 (seconds). Would be better if Hyatt used
>>a more precise timer I believe. Not sure what, but, this old one is the one I
>>always use... I'm sure he's got something better. Just incase, here you go...
>
>Ok, I see the problem.  For some reason, the bench command is calculating NPS
>with a time granularity of seconds, rather than the centiseconds used elsewhere
>in the program.  Maybe I'll hack a fix for this and send it to Bob.


That's possible, since I didn't write the bench code.  Not sure why it would
be done in second granularity other than to perhaps forget about the 1/100ths
for simplicity in displaying the output.



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.