Stockfish default bench can be easily improved

[Jouni]

Now it search only to 13 plies depth, not enough. Whole bench takes 2,5 seconds. Better change either depth to about 18 or movetime to about 500ms and You get automatically usable result!

[syzygy]

Now it search only to 13 plies depth, not enough. Whole bench takes 2,5 seconds. Better change either depth to about 18 or movetime to about 500ms and You get automatically usable result!

Usable how?

[Jouni]

Default bench gives too low values:
bench
Nodes/second : 1763442

but with
bench 64 1 20 default depth
Nodes/second : 2005502

[BeyondCritics]

You are giving more hash memory, so it should be faster, but i believe this is not of concern to them.
A question which is _really_ of concern to them could be: What testing parameters should i choose, to test a none functional changed version of one stockfish version against an older version, to reach a conclusion of given quality in the least amount of time? Figure it out for them and this will really help.

[MikeB]

Now it search only to 13 plies depth, not enough. Whole bench takes 2,5 seconds. Better change either depth to about 18 or movetime to about 500ms and You get automatically usable result!

For a speed test change, a search depth of 30 is often used.

The short bench is simply used as a "bench signature" by the SF developers 99% of the time. Total bench nodes should not change when a non functional simplification is introduced. Occasionally, a search depth of 13 is not long enough , and a deeper search is required to identify a potential non functional simplification that is, in reality, a functional simplification, but that is usually not the case. For the purpose it was designed for, it is adequate, ymmv.