Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Calculation mysteries: Rybka Benchmarks (Aaron Gordon)

Author: Vasik Rajlich

Date: 15:40:45 12/10/05

Go up one level in this thread


On December 09, 2005 at 16:16:39, Ernest Bonnem wrote:

>In  http://www.talkchess.com/forums/1/message.html?467780
>Aaron Gordon tried to look at the influence of L2 cache on the nps speed of
>Rybka.
>
>But the way the UCI launch (to depth 14) leads to numbers is strange, in all
>results reported.
>
>On my AMD mobile 2000+ (1.66 GHz, 32bit, L2=256k), the number of nodes attains
>the correct value, 3988843, and the "time" in milliseconds ranged from 53328 to
>54259 (in the 10+ tries I made).
>But my nps was always the same: 76708,
>which actually corresponds to a time of 52.000 seconds!
>Strange rounding of 53.328 to 54.259 !!!
>
>So the accuracy of the nps calculation is , to say the least, relative,
>particularly for fast machines (in the case of Aaron, the time of the 64bit
>Rybka on his fastest machine was around 15 sec)
>
>Maybe some experts have an understanding and a way to prevent such inaccuracies?

Hi,

this will be fixed in the next release.

It is due to the fact that Rybka divides nodes by seconds rather than
milliseconds for t > 10 seconds (IIRC).

Vas



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.