Layman question on Stockfish bench

[mohzus]

Hello people,
I would like to know exactly how the "bench" command works for Stockfish. In the code I only find the word "bench" at 2 places but it is not clear to me what it does exactly. Bear in mind that I don't know programming.

I find it really amazing that the nodes searched will differ according the code differs even in the slightest way. I hope to understand what's going on.

Also, is it possible that two different versions of SF produce the same number of nodes searched from the bench command? (i.e. that the signature is the same for 2 different codes). My guess is that the probability is extremely small but greater than 0 nonetheless.

Thank you for any explanation.

[mcostalba]

Thank you for any explanation.

It searches 30 positions up to depth 13


P.S: Almost everything has non-zero probability, even that <name a very weak program> wins TCEC.

[mohzus]

Thank you for any explanation.

It searches 30 positions up to depth 13


P.S: Almost everything has non-zero probability, even that <name a very weak program> wins TCEC.

Thank you Marco.
Just to make sure I've understood well, is the following correct?: If I make a modification in the code and the signature is less than the original code, then I can safely assume that the modified code is "faster" in the sense that in average SF will reach depth 13 with less nodes compared to the original code.

[mcostalba]

Thank you for any explanation.

It searches 30 positions up to depth 13


P.S: Almost everything has non-zero probability, even that <name a very weak program> wins TCEC.

Thank you Marco.
Just to make sure I've understood well, is the following correct?: If I make a modification in the code and the signature is less than the original code, then I can safely assume that the modified code is "faster" in the sense that in average SF will reach depth 13 with less nodes compared to the original code.

On those positions yes, but nothing you can say in general case. Perhaps on another set of positions searched nodes is higher. You have to assume signature is mostly a random number that is unique to that specific version, and this is the way it is used: we don't make any consideration about search features out of bench number.

[mohzus]

I see, Marco. Thank you for the explanation.