cutechess-cli concurrency

[zenpawn]

Any advice on the best concurrency setting to use on a standard i7 laptop such that a gauntlet can be played in the least amount of time without affecting the games? I've been using 4 for 40/30+1 time controls, and am just now venturing 6 for 40/60+1 games.

[Ferdy]

Any advice on the best concurrency setting to use on a standard i7 laptop such that a gauntlet can be played in the least amount of time without affecting the games? I've been using 4 for 40/30+1 time controls, and am just now venturing 6 for 40/60+1 games.

How many cores does the processor of your laptop has? Does the processor supports hyper-threading?

[zenpawn]

How many cores does the processor of your laptop has? Does the processor supports hyper-threading?

It's a quad-core (i7-4700MQ @ 2.40GHz) and has hyperthreading enabled.

[Ferdy]

How many cores does the processor of your laptop has? Does the processor supports hyper-threading?

It's a quad-core (i7-4700MQ @ 2.40GHz) and has hyperthreading enabled.

You may try 6 or 7. But important is try to run same engine on the TC you like to test and see if you can get 50% or close to it after 1000 games. If you cannot get 50% or close to it then there could be an issue in hyper-threading, in that case you may try affinity of engine to a thread. Say engine1 and engine2 can only use threads 0 to 5, that is using cores 1, 2 and 3. Core1=threads 0 and 1, core2=threads 2 and 3, core3=threads 4 and 5. Core4=threads 6 and 7 are free from engine use. Try this setup using same engine, see if you can get 50% or close to it after some games.

[zenpawn]

You may try 6 or 7. But important is try to run same engine on the TC you like to test and see if you can get 50% or close to it after 1000 games. If you cannot get 50% or close to it then there could be an issue in hyper-threading, in that case you may try affinity of engine to a thread. Say engine1 and engine2 can only use threads 0 to 5, that is using cores 1, 2 and 3. Core1=threads 0 and 1, core2=threads 2 and 3, core3=threads 4 and 5. Core4=threads 6 and 7 are free from engine use. Try this setup using same engine, see if you can get 50% or close to it after some games.

Thank you for this way to determine whether concurrency might be set too high. Regarding the affinity, I'm not seeing such options in the cutechess-cli documentation.

[Ferdy]

You may try 6 or 7. But important is try to run same engine on the TC you like to test and see if you can get 50% or close to it after 1000 games. If you cannot get 50% or close to it then there could be an issue in hyper-threading, in that case you may try affinity of engine to a thread. Say engine1 and engine2 can only use threads 0 to 5, that is using cores 1, 2 and 3. Core1=threads 0 and 1, core2=threads 2 and 3, core3=threads 4 and 5. Core4=threads 6 and 7 are free from engine use. Try this setup using same engine, see if you can get 50% or close to it after some games.

Thank you for this way to determine whether concurrency might be set too high. Regarding the affinity, I'm not seeing such options in the cutechess-cli documentation.

Search for cpu affinity on web.

Cutechess does not handle it.

I tried process lasso before.

[Nordlandia]

Process Laso let you select CPU affinity.

I recommend Process Laso by Bitsum

I use it on both laptops.

Erin - one question: Why do you use Cutechess-Cli when 64-bit version is available with 6-men adjudication.

https://github.com/cutechess/cutechess/ ... -304545748

[zenpawn]

Erin - one question: Why do you use Cutechess-Cli when 64-bit version is available with 6-men adjudication.

https://github.com/cutechess/cutechess/ ... -304545748

I use the command line so there's no GUI involvement; I've noticed strange delays in Arena. Also, it helps my sanity to not observe the game in progress.

That said, the link you provided also includes a CLI in the zip. Perhaps I'm not on the latest version (though I never mentioned a version number), so I'll check it out. I don't bother with adjudication as my engine is still quite capable of losing/drawing a won endgame.