Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Coparing two Identical Programs using Different Processors Speed !

Author: Uri Blass

Date: 09:40:52 01/29/01

Go up one level in this thread


On January 29, 2001 at 12:21:25, Severi Salminen wrote:

>>If you do the right comparison then it is clear that B can get better results
>>with faster hardware.
>>
>>You only need to do the right comparison and use faster time control on the
>>faster hardware so A will calculate the same number of nps per move when B will
>>calculate more nps per move.
>>
>>If we agree that for all programs more nps per move is better than it is clear
>>that B can get better result on faster hardware when you use the right time
>>control that is different for both hardwares.
>
>Bottom line: of course more NPS is better, but the ratio of NPS on different
>platforms doesn't necessarely have a relation to the ratio of playing strenghts.
>It has if the branching factor is the same. But usually it is not, so you can't
>compare. But it _is_ true if we scale the time control down. So we have two
>cases:
>
>1. gain from speed (this _is_ the same as longer time controls on same platform,
>if both programs use same pondering). Many things have incluence on this:
>branching factor, hashtable management...
>
>2. gain from different architecture (this is something you just have to test).

I think that 2 is the main reason for difference in results on the same time
control on different platforms

The number of Gandalf's nodes per second channges from 1/6 of Junior nodes to
almost 1/2 of Junior nodes from changing the platform based on a previous post.

I believe that there is no big difference in the branching factor of the top
programs.
It is also impossible to use the branching factor to know which program is
better because it is possible to reduce the branching factor by doing more
pruning and it is not going to do the program stronger.

I think that there may be big difference if you compare blitz results and
tournament time control results on the same hardware but increasing the time of
both programs by a factor of 2 is not going to do a big difference and it is the
reason that I believe that architecture is dominant when you compare the same
time control when the hardware is different.

Uri



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.