Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Why bother with NPS??

Author: Ricardo Gibert

Date: 05:25:51 09/10/03

Go up one level in this thread


On September 09, 2003 at 23:47:20, Christophe Theron wrote:

>On September 09, 2003 at 09:01:57, Arshad Syed wrote:
>
>>I read in a post here that tweaking NPS a bit here and there wouldn't really
>>matter since the node depth wouldn't increase with a marginal increase in NPS.
>>The NPS would have to drastically increase in order to increase the depth by a
>>level, at depths > 6.
>>
>>Please give me your opinion on this, since I don't want to be tweaking code
>>unnecessarily, especially if it is not going to make a difference.
>>
>>Thanks,
>>Arshad
>
>
>
>Doubling the NPS without changing the rest of the program typically means a 60
>to 70 elo points improvement (70 was the average for SSDF hardware from a few
>years ago, but I tend to believe that with faster hardware the gain is less -
>that remains to be seen).
>
>So 100% faster -> +65 elo points
>    50% faster -> +33 elo points (half of 65)
>    25% faster -> +16 elo points
>...and so on.


This should be:

100% faster -> +65 elo points
 50% faster -> +38 elo points (*not* half of 65)
 25% faster -> +21 elo points

See Martin's post.

>
>You do not need to think in term of ply increase from NPS increase: it does not
>make much sense for small NPS gains.
>
>But you can think in term of elo gain, even for the smallest speed improvements.
>This tells you more than the gain in ply depth in my opinion.
>
>
>
>    Christophe



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.