COLLECTION: Which engines comes with an own bench test?

[Frank Quisinsky]

Hi Volker,

thats interesting.
On weekend I will try out your latest Herman version. Also the bench test

Will used my notebook for it.

Best
Frank

[CRoberson]

Hi Charles,

I used for the information I got from Ares 1004 only the np/s. Hope this is right.

Thanks

Best
Frank

Hi Frank,

After typing Ares1.004.exe -bench -fd 15

you should get something like this (not exactly as I am running a newer version and a fast machine).

Code: Select all

Running benchmark 
    Time Interior Nodes Quiesce Nodes Total Nodes        NPS
   83800    77094665      78355286     155449951     1855000
Leaf Interior Nodes searched = 2756792
Transposition Table Hits 3663441  Misses  25637230  Ratio = 12
Pawn misses 7995735 Pawn attempts 101648232 Pawn miss rate = 7.86608

The interpretation is as folllows:
The test positions ran for a depth of 15 ply each
The total time was 83800 milliseconds = 83.8 seconds
Interior nodes are the nodes of the main tree without qsearch for all positions total
Quiesce Nodes = qnodes for all positions total
Total Nodes = Sum total of all nodes (interior and quiesce) for all positions
NPS = 1.85 Million Nodes per second

The rest should be obvious except for leaf interior nodes. This is the leaf nodes of the main search without qsearch. These nodes are the nodes that the search ends on before calling the qsearch.

Regards,
Charles

[bob]

NoonianChess and Ares both have their own bench test which can be started at a command prompt.

NoonianChess -bench -fd depth
Ares -bench -fd depth

I wouldn't run NoonianChess any deeper than 10 ply; it is too slow. Depending on the speed of the computer, you could run Ares 14 ply deep or more.

Ares -bench -fd 12 // searches a set of positions 12 ply deep.

Ares reports several things including NPS and time. Time is reported in milliseconds, so 12640 is 12.64 seconds. Ares with -fd 10 takes 12.64 seconds on my laptop (Intel Core 2 Duo T5500) in 32 bit mode.

Having an option reduces the usefulness. Different depths produce different times, making things hard to compare. I used internal fixed-depth searches on the benchmark positions Crafty uses...

[CRoberson]

NoonianChess and Ares both have their own bench test which can be started at a command prompt.

NoonianChess -bench -fd depth
Ares -bench -fd depth

I wouldn't run NoonianChess any deeper than 10 ply; it is too slow. Depending on the speed of the computer, you could run Ares 14 ply deep or more.

Ares -bench -fd 12 // searches a set of positions 12 ply deep.

Ares reports several things including NPS and time. Time is reported in milliseconds, so 12640 is 12.64 seconds. Ares with -fd 10 takes 12.64 seconds on my laptop (Intel Core 2 Duo T5500) in 32 bit mode.

Having an option reduces the usefulness. Different depths produce different times, making things hard to compare. I used internal fixed-depth searches on the benchmark positions Crafty uses...

That is partially true. It depends on the intended purpose. I didn't intend the benchmark for general CPU testing and for that you are correct. I put it in NoonianChess and then in Ares (more elaborate in Ares), as a development tool. As a development tool, it makes sense to allow for multiple depths. It answered a number of conundrums for me.