Stockfish bench surprise

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

Jouni
Posts: 3727
Joined: Wed Mar 08, 2006 8:15 pm
Full name: Jouni Uski

Stockfish bench surprise

Post by Jouni »

This is SF 17 default bench with 1 and 6 cores:

bench 512 1 25

Total time (ms) : 104096
Nodes searched : 93675716
Nodes/second : 899897

bench 512 6 25

Total time (ms) : 113035
Nodes searched : 488314553
Nodes/second : 4320029

So 1 core is faster :o . Is this meaning either 1) 6 core search does 5x more careful search or 2) "depth" is useless information?
Jouni
User avatar
Ras
Posts: 2713
Joined: Tue Aug 30, 2016 8:19 pm
Full name: Rasmus Althoff

Re: Stockfish bench surprise

Post by Ras »

Jouni wrote: Mon Nov 04, 2024 12:35 pmIs this meaning either 1) 6 core search does 5x more careful search or 2) "depth" is useless information?
Or 3) this is an NPS benchmark, not a actual search, and "depth" is for controlling how long the benchmark should take.
Rasmus Althoff
https://www.ct800.net
Jouni
Posts: 3727
Joined: Wed Mar 08, 2006 8:15 pm
Full name: Jouni Uski

Re: Stockfish bench surprise

Post by Jouni »

Addition: older versions of SF don't behave this way in bench.
Jouni
Viz
Posts: 223
Joined: Tue Apr 09, 2024 6:24 am
Full name: Michael Chaly

Re: Stockfish bench surprise

Post by Viz »

depth always was a semi-useless information since the implementation of lazySMP