Leela Chess Zero 42565 vs Stockfish 140619

Discussion of computer chess matches and engine tournaments.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
mwyoung
Posts: 1641
Joined: Wed May 12, 2010 8:00 pm

Re: Leela Chess Zero 42565 vs Stockfish 140619

Post by mwyoung » Sat Jun 15, 2019 5:33 pm

Nordlandia wrote:
Sat Jun 15, 2019 5:20 pm
mwyoung: the math does not go up. Stockfish make claim of 100% cpu by itself. There is clearly performance interference whether you can notice it or not. Leela need 1-2 threads to feed the gpu. Those threads messes the game up because they're already used by stockfish. That being said, the streaming recording device make also claim of cpu time so leaving 1-thread free for the OS make total sense. SF=29 threads - Lc0=2 threads and 1 thread free for the operating system for free flow/background tasks and no slowdown.

Graham Banks said this on chess2u few years ago.

"I should add that each engine should access tablebases from a different drive in order to avoid clashes on the hard drive"

It's a good point, i'll do it. I have 2 SSDs, so i'll place 5-men syzygy on each and give each engine seperate directory for syzygy.

Look at this thread.

https://chess.stackexchange.com/questio ... nent-brain

And this one

http://www.chess2u.com/t4335-ponder-on-or-off
You are correct the match does not go up. But you are not asking the right question.

What is the cost and what is the return.

Hint: Lc0 is not a AB engine.

Your method you are suggesting is the worst way to leverage a multi core AMD CPU.
Professing themselves to be wise, they became fools,
Take on me. foes 0

mwyoung
Posts: 1641
Joined: Wed May 12, 2010 8:00 pm

Re: Leela Chess Zero 42565 vs Stockfish 140619

Post by mwyoung » Sat Jun 15, 2019 6:10 pm

mwyoung wrote:
Sat Jun 15, 2019 5:33 pm
Nordlandia wrote:
Sat Jun 15, 2019 5:20 pm
mwyoung: the math does not go up. Stockfish make claim of 100% cpu by itself. There is clearly performance interference whether you can notice it or not. Leela need 1-2 threads to feed the gpu. Those threads messes the game up because they're already used by stockfish. That being said, the streaming recording device make also claim of cpu time so leaving 1-thread free for the OS make total sense. SF=29 threads - Lc0=2 threads and 1 thread free for the operating system for free flow/background tasks and no slowdown.

Graham Banks said this on chess2u few years ago.

"I should add that each engine should access tablebases from a different drive in order to avoid clashes on the hard drive"

It's a good point, i'll do it. I have 2 SSDs, so i'll place 5-men syzygy on each and give each engine seperate directory for syzygy.

Look at this thread.

https://chess.stackexchange.com/questio ... nent-brain

And this one

http://www.chess2u.com/t4335-ponder-on-or-off
You are correct the math does not go up. But you are not asking the right question.

What is the cost and what is the return.

Hint: Lc0 is not a AB engine.

Your method you are suggesting is the worst way to leverage a multi core AMD CPU.
Professing themselves to be wise, they became fools,
Take on me. foes 0

User avatar
Nordlandia
Posts: 2447
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: Leela Chess Zero 42565 vs Stockfish 140619

Post by Nordlandia » Sat Jun 15, 2019 6:40 pm

What is the cost and what is the return.

In chess engine terms, playing with ponder on put a lot of strain on the system resources. Leela need 1-2 threads itself and many recommend one thread for the OS. Tweaking SF threads to 29 from 32 it's not the end of the world. It's a necessity for stability reasons. As previously stated, leela need CPU time and that CPU time is already being used on stockfish. So it messes up the games.

mwyoung
Posts: 1641
Joined: Wed May 12, 2010 8:00 pm

Re: Leela Chess Zero 42565 vs Stockfish 140619

Post by mwyoung » Sat Jun 15, 2019 6:54 pm

Nordlandia wrote:
Sat Jun 15, 2019 6:40 pm
What is the cost and what is the return.

In chess engine terms, playing with ponder on put a lot of strain on the system resources. Leela need 1-2 threads itself and many recommend one thread for the OS. Tweaking SF threads to 29 from 32 it's not the end of the world. It's a necessity for stability reasons. As previously stated, leela need CPU time and that CPU time is already being used on stockfish. So it messes up the games.
Again that is the worst way to leverage the CPU. You see my system pondering. You can see it play with ponder off. Hmmm...

What issue do you see?

Not all workloads are quantized into whole threads.

Don't believe your lying eyes.

And please don't project your ignorance onto me.
Professing themselves to be wise, they became fools,
Take on me. foes 0

MikeB
Posts: 3461
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: Leela Chess Zero 42565 vs Stockfish 140619

Post by MikeB » Sun Jun 16, 2019 12:39 am

mwyoung wrote:
Sat Jun 15, 2019 6:54 pm
Nordlandia wrote:
Sat Jun 15, 2019 6:40 pm
What is the cost and what is the return.

In chess engine terms, playing with ponder on put a lot of strain on the system resources. Leela need 1-2 threads itself and many recommend one thread for the OS. Tweaking SF threads to 29 from 32 it's not the end of the world. It's a necessity for stability reasons. As previously stated, leela need CPU time and that CPU time is already being used on stockfish. So it messes up the games.
Again that is the worst way to leverage the CPU. You see my system pondering. You can see it play with ponder off. Hmmm...

What issue do you see?

Not all workloads are quantized into whole threads.

Don't believe your lying eyes.

And please don't project your ignorance onto me.
The prognosis is simple, the match you have set up has a request for more than 100% of the resources of the entire match. You may run your little match and everything looks fine to your eyes. but if you were to try to duplicates your results over and over again - lots of luck with that. You are creating, so much noise with overloading your setup, that your match results are untrustworthy. Fishtest sees this all the time, that is why fishttest is run with ponder off and the threads being run are less than the total number of cores. Even with the setup , they see statistical anomalies , it's about getting consistent repeatable results and when you design your match tfor it to be impossible to obtain consistent repeatable results, your results are worthless. Now it's true we do not know actually whether or not you are getting consistent repeatable results, but most of us have been around long enough, that with your described setup as YOU described to us , you will be unable to obtain consistent repeatable results. You settings are designed to introduce the maximum amount of noise as possible. You don't want to listen to anyone , which is fine, but don't expect a lot of love in return.

mwyoung
Posts: 1641
Joined: Wed May 12, 2010 8:00 pm

Re: Leela Chess Zero 42565 vs Stockfish 140619

Post by mwyoung » Sun Jun 16, 2019 1:48 am

MikeB wrote:
Sun Jun 16, 2019 12:39 am
mwyoung wrote:
Sat Jun 15, 2019 6:54 pm
Nordlandia wrote:
Sat Jun 15, 2019 6:40 pm
What is the cost and what is the return.

In chess engine terms, playing with ponder on put a lot of strain on the system resources. Leela need 1-2 threads itself and many recommend one thread for the OS. Tweaking SF threads to 29 from 32 it's not the end of the world. It's a necessity for stability reasons. As previously stated, leela need CPU time and that CPU time is already being used on stockfish. So it messes up the games.
Again that is the worst way to leverage the CPU. You see my system pondering. You can see it play with ponder off. Hmmm...

What issue do you see?

Not all workloads are quantized into whole threads.

Don't believe your lying eyes.

And please don't project your ignorance onto me.
The prognosis is simple, the match you have set up has a request for more than 100% of the resources of the entire match. You may run your little match and everything looks fine to your eyes. but if you were to try to duplicates your results over and over again - lots of luck with that. You are creating, so much noise with overloading your setup, that your match results are untrustworthy. Fishtest sees this all the time, that is why fishttest is run with ponder off and the threads being run are less than the total number of cores. Even with the setup , they see statistical anomalies , it's about getting consistent repeatable results and when you design your match tfor it to be impossible to obtain consistent repeatable results, your results are worthless. Now it's true we do not know actually whether or not you are getting consistent repeatable results, but most of us have been around long enough, that with your described setup as YOU described to us , you will be unable to obtain consistent repeatable results. You settings are designed to introduce the maximum amount of noise as possible. You don't want to listen to anyone , which is fine, but don't expect a lot of love in return.
This is not the first match. I just finished a 285 game ponder on match. I have played ponder on matches many times.

The match results are consistent with ponder off matches.

So not only are you wrong. Anyone can view the tapes and see you are wrong.
Professing themselves to be wise, they became fools,
Take on me. foes 0

mwyoung
Posts: 1641
Joined: Wed May 12, 2010 8:00 pm

Re: Leela Chess Zero 42565 vs Stockfish 140619

Post by mwyoung » Sun Jun 16, 2019 1:49 am

MikeB wrote:
Sun Jun 16, 2019 12:39 am
mwyoung wrote:
Sat Jun 15, 2019 6:54 pm
Nordlandia wrote:
Sat Jun 15, 2019 6:40 pm
What is the cost and what is the return.

In chess engine terms, playing with ponder on put a lot of strain on the system resources. Leela need 1-2 threads itself and many recommend one thread for the OS. Tweaking SF threads to 29 from 32 it's not the end of the world. It's a necessity for stability reasons. As previously stated, leela need CPU time and that CPU time is already being used on stockfish. So it messes up the games.
Again that is the worst way to leverage the CPU. You see my system pondering. You can see it play with ponder off. Hmmm...

What issue do you see?

Not all workloads are quantized into whole threads.

Don't believe your lying eyes.

And please don't project your ignorance onto me.
The prognosis is simple, the match you have set up has a request for more than 100% of the resources of the entire match. You may run your little match and everything looks fine to your eyes. but if you were to try to duplicates your results over and over again - lots of luck with that. You are creating, so much noise with overloading your setup, that your match results are untrustworthy. Fishtest sees this all the time, that is why fishttest is run with ponder off and the threads being run are less than the total number of cores. Even with the setup , they see statistical anomalies , it's about getting consistent repeatable results and when you design your match tfor it to be impossible to obtain consistent repeatable results, your results are worthless. Now it's true we do not know actually whether or not you are getting consistent repeatable results, but most of us have been around long enough, that with your described setup as YOU described to us , you will be unable to obtain consistent repeatable results. You settings are designed to introduce the maximum amount of noise as possible. You don't want to listen to anyone , which is fine, but don't expect a lot of love in return.
This is not the first match. I just finished a 285 game ponder on match. I have played ponder on matches many times.

The match results are consistent with ponder off matches.

So not only are you wrong. Anyone can view the tapes and see you are wrong.

You geniuses need to stop thinking of Lc0 as a AB engine. And stop thinking threads are the same as utilization.
Professing themselves to be wise, they became fools,
Take on me. foes 0

User avatar
Nordlandia
Posts: 2447
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: Leela Chess Zero 42565 vs Stockfish 140619

Post by Nordlandia » Sun Jun 16, 2019 3:19 pm

It's still an exercise in futility with your current settings.

mwyoung
Posts: 1641
Joined: Wed May 12, 2010 8:00 pm

Re: Leela Chess Zero 42565 vs Stockfish 140619

Post by mwyoung » Sun Jun 16, 2019 3:34 pm

Nordlandia wrote:
Sun Jun 16, 2019 3:19 pm
It's still an exercise in futility with your current settings.
And you just keep on trying to impressing me with your ignorance. Since you don't know what my setting are, or did anyone ask before being attacked. Or why this method works with Lc0 and not AB engines. But this is to be expected in the troll farm this forum has become. As my ponder on matches play on flawlessly.
Professing themselves to be wise, they became fools,
Take on me. foes 0

User avatar
Laskos
Posts: 9447
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: Leela Chess Zero 42565 vs Stockfish 140619

Post by Laskos » Sun Jun 16, 2019 4:36 pm

mwyoung wrote:
Sun Jun 16, 2019 3:34 pm
Nordlandia wrote:
Sun Jun 16, 2019 3:19 pm
It's still an exercise in futility with your current settings.
And you just keep on trying to impressing me with your ignorance. Since you don't know what my setting are, or did anyone ask before being attacked. Or why this method works with Lc0 and not AB engines. But this is to be expected in the troll farm this forum has become. As my ponder on matches play on flawlessly.
Someone here is a vivid example that this forum had a good quantity of absolute imbeciles since its beginnings.

Post Reply