pondering

[bezzy3004]

When playing two engines against each other at tournament level whuch is best pondeing on or off? my cpu is amd64x2 5000+ 2.6ghz with on they each use 50 percent is that correct would they be hindered with pondeing off?

[Graham Banks]

When playing two engines against each other at tournament level whuch is best pondeing on or off? my cpu is amd64x2 5000+ 2.6ghz with on they each use 50 percent is that correct would they be hindered with pondeing off?

If you only have 1CPU, best use ponder off, but as you have 2CPU you can use ponder on as long as you don't use the computer for other tasks simultaneously.

[hgm]

Ponder on will not hurt on a dual. But engines do not benifit as much from pondering as they would from giving them twice longer TC with ponder off, and then playing two games simultaneously.

[IWB]

When playing two engines against each other at tournament level whuch is best pondeing on or off? my cpu is amd64x2 5000+ 2.6ghz with on they each use 50 percent is that correct would they be hindered with pondeing off?

If you only have 1CPU, best use ponder off, but as you have 2CPU you can use ponder on as long as you don't use the computer for other tasks simultaneously.

... additionaly "Ponder ON" is simulating a REAL tournament (online or on the board) while "Ponder OFF" is artificial and not intended by most programmers!

Bye
Ingo

[hgm]

OTOH, playing with ponder OFF is simulating a REAL rating measurement, while Ponder ON is artificial and not intended by most programmers (whose engines can often not ponder at all...)

It escapes me why anyone would want to 'simulate' a real tournament, when not actually participating in one. I, for one, am only interested in getting the best-quality Chess play for the available CPU power.

[IWB]

Hello

OTOH, playing with ponder OFF is simulating a REAL rating measurement, while Ponder ON is articicial and not intended by most programmers (whose engines can oftennot oponder at all...)

Sorry I have some objections about this.

1. There are no tournament by humans where the humans stops thinking while the opponent is to move, therefore PONDER ON is the normal mode for a tournament by definition
2. It is true that rating list are often made with ponder off for historical reasons (single core CPUs), so many rating measurements are done this way - but that does not mean that it is better or more correct than with ponder on. The opposite might be the case (see Argument 1).
3. To my knowledge MOST Engines (at least at a certain development stage), and for sure ALL comercial ones, can ponder of course!

So different views of the same object here

Bye
Ingo

[hgm]

Not all of this can described as 'views', some of it is more like 'facts'.

*) In lowest approximation 0% of all engines is commercial. So what they do or don't is not very relevant for any discussion on Chess engines...

*) How many of the 464 engines participating in Chess War XIII to your knowledge support pondering?

* Computers are not Humans.

There are also no tournements where a Human searches more than a few hundred nodes per move. So are you also in favor of putting a limit of 200 nodes on the engine search, to "simulate a REAL tournament"? Requiring that computers simulate Humans in every respect other than pondering quickly leads to totally absurd situations. So no one even considers it. Why make an exception for pondering? What Humans do or don't carries even less weight than what commercial programs do.

[IWB]

Hello

Not all of this can described as 'views', some of it is more like 'facts'.

*) In lowest approximation 0% of all engines is commercial. So what they do or don't is not very relevant for any discussion on Chess engines...

But 99.x% of all engine users uses this 0%!

*) How many of the 464 engines participating in Chess War XIII to your knowledge support pondering?

How many engine users world wide have heard of "chess War XXX whatever..." and how many engine users world wide really USES engines which cant ponder?

* Computers are not Humans.

Very true, but is not changing my argument about chess tournaments and their relevance!

There are also no tournements where a Human searches more than a few hundred nodes per move. So are you also in favor of putting a limit of 200 nodes on the engine search, to "simulate a REAL tournament"? ...

You are cherry picking - "I" think that the vast majority of chess players and a very good percentage of the people who know chess engines (must of them end their knowledge after the term "Fritz") dont mind in computer nodes vs human nodes - and btw the 200 nodes top humans are not doing that bad until just recently vs Engines!
When was the last tournament where a programmer took part which was held in "ponder off"? WC, ICC tourneys, Leiden, Paderborn ... whatever, as soon as a tourney is played with programmers they use "Ponder ON" if the engine can do that!
Playing touneys ponder off is a makeshift as tere where/are just single core CPUs and some guys wanted to make rating lists with a lot of games.
I repaet and stand for it: "A real tournament should be ponder on"!

Bye and regards
Ingo

[kranium]

i think it depends on the processor...
and if it supports hyperthreading or not (and if if hyperthreading is enabled or not)...
if hyperthreading, 2 (physical) CPUs appear to windows as 4 logical processors...

for ex: on my Intel Xeon 2 CPU (with hyperthreading) each thread = 25%
thus if ponder=true, and threads=1, the machine can easily handle
engine vs engine, and still maintain 50% CPU free for other tasks...

on some other processors, 1 thread = 50% CPU
so if ponder=true then 100% CPU state is often reached, and the PC will be hard pressed to perform any other tasks...

i have an older model pentium IV, and 1 thread = 100% CPU, so 100% CPU is reached even with pondering off...
and with pondering enabled, the competition for CPU time is quite severe.

i also have a newer model of Pentium IV (with hypethreading),
and in this case 1 thread = 50% CPU...

Norm

[hgm]

But 99.x% of all engine users uses this 0%!

So your statements pertain only to commercial engines, and have no general relevance. That is all I wanted to point out.

When was the last tournament where a programmer took part which was held in "ponder off"? WC, ICC tourneys, Leiden, Paderborn ... whatever, as soon as a tourney is played with programmers they use "Ponder ON" if the engine can do that!
Playing touneys ponder off is a makeshift as tere where/are just single core CPUs and some guys wanted to make rating lists with a lot of games.
I repaet and stand for it: "A real tournament should be ponder on"!

Again your statement is heavily colored by you very limited focus on toy computers and the commercial programs written for them.

If a program written for a 1024-core supercomputer (i.e. not a cluster, but the real thing) would participate in such a tournament, and they would have to hire time on such a facility, they would almost certainly prefer to spend the available budget of CPU-seconds by running as many CPUs as they can afford when their own clock is running, rather than wasting CPU seconds on pondering. Better to run 512 CPUs during half the time, than 256 CPUs all the time.

From the viewpoint of efficient use of hardware, pondering is simply a bad idea.