...and the choices are:
1: 3 cores ponder OFF
2: 1 core ponder ON
The computer that is going to be used is an Intel Core i7 3770k @ 4400 MHz.
With ponder OFF the engines will search faster/deeper with 3 cores as opposed to just 1.
On the other hand, with ponder ON it doesn't matter if the engine is SMP capable or not as all engines have equal terms.
What do you prefer?
TCEC resurrection - need to decide between ponder ON or OFF
Moderators: hgm, Rebel, chrisw
-
- Posts: 1833
- Joined: Thu Jun 22, 2006 12:07 am
-
- Posts: 1154
- Joined: Fri Jun 23, 2006 5:18 am
Re: TCEC resurrection - need to decide between ponder ON or
My opinion on this matter has recently changed.
-Sam
-Sam
-
- Posts: 41468
- Joined: Sun Feb 26, 2006 10:52 am
- Location: Auckland, NZ
Re: TCEC resurrection - need to decide between ponder ON or
Hi Martin,Martin Thoresen wrote:...and the choices are:
1: 3 cores ponder OFF
2: 1 core ponder ON
The computer that is going to be used is an Intel Core i7 3770k @ 4400 MHz.
With ponder OFF the engines will search faster/deeper with 3 cores as opposed to just 1.
On the other hand, with ponder ON it doesn't matter if the engine is SMP capable or not as all engines have equal terms.
What do you prefer?
it's great to see you back on the scene.
For what it's worth, I think you should go with your own preference. You'll enjoy things more that way.
Graham.
gbanksnz at gmail.com
-
- Posts: 782
- Joined: Tue Dec 01, 2009 11:10 am
Re: TCEC resurrection - need to decide between ponder ON or
1++Graham Banks wrote:Hi Martin,Martin Thoresen wrote:...and the choices are:
1: 3 cores ponder OFF
2: 1 core ponder ON
The computer that is going to be used is an Intel Core i7 3770k @ 4400 MHz.
With ponder OFF the engines will search faster/deeper with 3 cores as opposed to just 1.
On the other hand, with ponder ON it doesn't matter if the engine is SMP capable or not as all engines have equal terms.
What do you prefer?
it's great to see you back on the scene.
For what it's worth, I think you should go with your own preference. You'll enjoy things more that way.
Graham.
-
- Posts: 3232
- Joined: Mon May 31, 2010 1:29 pm
- Full name: lucasart
Re: TCEC resurrection - need to decide between ponder ON or
Ponder OFF, without hesitation!
IMO, the best way to conduct a tournament is as follows (assuming you have N CPU, example N=8 on my i7)
=> non SMP engines
- run N-1 games in parrallel (using cutechess-cli for example)
- pondering OFF
- each engine uses one search thread
- you leave 1 core for the operating system (even when it "looks" idle, it never is, especially if you are using windows). this is uber-important!
=> SMP engines
- One game at a time
- Each engine has pondering OFF, and uses N-1 search threads (or maybe N, but remember that you're introducing another source of variance that you cannot measure, so you can't trust blindly the LOS from any models etc.)
In no event does pondering make any sense. And I really don't understand why CEGT is now doing another rating list with pondering ON for example.
The only case where pondering makes sense, is when you (human player) are playing against the computer, and want the computer to think on your time.
IMO, the best way to conduct a tournament is as follows (assuming you have N CPU, example N=8 on my i7)
=> non SMP engines
- run N-1 games in parrallel (using cutechess-cli for example)
- pondering OFF
- each engine uses one search thread
- you leave 1 core for the operating system (even when it "looks" idle, it never is, especially if you are using windows). this is uber-important!
=> SMP engines
- One game at a time
- Each engine has pondering OFF, and uses N-1 search threads (or maybe N, but remember that you're introducing another source of variance that you cannot measure, so you can't trust blindly the LOS from any models etc.)
In no event does pondering make any sense. And I really don't understand why CEGT is now doing another rating list with pondering ON for example.
The only case where pondering makes sense, is when you (human player) are playing against the computer, and want the computer to think on your time.
Theory and practice sometimes clash. And when that happens, theory loses. Every single time.
-
- Posts: 263
- Joined: Sun Oct 28, 2012 11:45 am
Re: TCEC resurrection - need to decide between ponder ON or
And why do you think that's so?lucasart wrote: In no event does pondering make any sense. And I really don't understand why CEGT is now doing another rating list with pondering ON for example.
There is a long diatribe on this point. Many say that PB is the best since engines are meant to be used that way (the time management etc. is meant for PB), other says that PB introduces a possible sharing of resources by the engines on 1 PC.
I have yet to understand why with PB ON the possibility of "sharing" should be more than with PB OFF.
I am not biased one way or another and in my tests the results are usually the same (apart rare exception) one way or another. Naturally if one has the leisure to have 2 same PC then it's neither a question. But even in this case many says that the "best" is 1 PC for the GUI + 2 for the engines to remove any possibility of sharing (yes, engine testers are paranoid, that's what I discovered).
In this particular case, IMO, it's more a diatribe if to have all engines have the same terms or not, more than PB ON/OFF. Either all 1 core or MP engines get 3 vs 1.
P.S: Then in your i7 you don't have 8 cores, but 4. If you use the logical cores for concurrency you would make a mess.
-
- Posts: 3232
- Joined: Mon May 31, 2010 1:29 pm
- Full name: lucasart
Re: TCEC resurrection - need to decide between ponder ON or
??????Lavir wrote: I have yet to understand why with PB ON the possibility of "sharing" should be more than with PB OFF.
It's obvious! If you have more searching threads than CPUs...
Theory and practice sometimes clash. And when that happens, theory loses. Every single time.
-
- Posts: 263
- Joined: Sun Oct 28, 2012 11:45 am
Re: TCEC resurrection - need to decide between ponder ON or
???????
And why should you have more searching threads than CPUs with PB ON????
And what it has to do with PB ON/OFF at all?????
On an i7 (for example) you can use 2 vs 2 PB ON or 4 vs 4 PB OFF, or, if you want to be use the PC, 1 vs 1 PB ON or 3 vs 3 PB OFF, so in both cases you naturally use the threads you have, not more.
If one is an idiot and gives 4 vs 4 PB ON (when you only have 4 threads, as in the above example), that's not a fault of PB ON, and sharing of resources has nothing to do with it.
And why should you have more searching threads than CPUs with PB ON????
And what it has to do with PB ON/OFF at all?????
On an i7 (for example) you can use 2 vs 2 PB ON or 4 vs 4 PB OFF, or, if you want to be use the PC, 1 vs 1 PB ON or 3 vs 3 PB OFF, so in both cases you naturally use the threads you have, not more.
If one is an idiot and gives 4 vs 4 PB ON (when you only have 4 threads, as in the above example), that's not a fault of PB ON, and sharing of resources has nothing to do with it.
-
- Posts: 3550
- Joined: Thu Jun 07, 2012 11:02 pm
Re: TCEC resurrection - need to decide between ponder ON or
My vote would be two cores with ponder Off. Three is a weird number.
-
- Posts: 3232
- Joined: Mon May 31, 2010 1:29 pm
- Full name: lucasart
Re: TCEC resurrection - need to decide between ponder ON or
But most testers are not programmers. It doesn't mean they are idiots, but it does mean that they are very likely to misuse/misunderstand threads/pondering. Just read the posts on here, and see for yourself.Lavir wrote: If one is an idiot and gives 4 vs 4 PB ON (when you only have 4 threads, as in the above example), that's not a fault of PB ON, and sharing of resources has nothing to do with it.
Anyway, I still think that pondering is a bad idea, except when you are playuing against the computer, and more generally when you don't have two engines sharing resources (example engine A running on computer A, vs engine B running on computer B, communicating via a server like freechess).
So, for fear of misuse, as well as lazyness, I never implemented the pondering option in my engine.
Theory and practice sometimes clash. And when that happens, theory loses. Every single time.