What you could read in Rybka 3 FAQ by Vas

[Rolf]

look here, I guess you can all read without membership.

http://rybkaforum.net/cgi-bin/rybkaforu ... ?pid=87852

Here you will learn how the KN/sec is done and meant. Also about the differences between Rybka 2 and 3. Hope this helps.

Could you please take a look, Christophe?

[Tony Thomas]

look here, I guess you can all read without membership.

http://rybkaforum.net/cgi-bin/rybkaforu ... ?pid=87852

Here you will learn how the KN/sec is done and meant. Also about the differences between Rybka 2 and 3. Hope this helps.

Could you please take a look, Christophe?

It seems to me that you dont even understand the point of the argument. The following quote is the only thing that I saw in the readme that's somewhat relevant, but that only explains the SMP scaling..From looking at Vas's second reply, it looks to me like he is just being stubborn and he is rather evasive in answering a direct question.

(*) Rybka's kn/s figure doesn't increase much when she runs on machines with more processors, compared to other programs. Does this mean that she scales worse on such hardware?

No. Rybka's kn/s on multi-processor machines reflects the estimated multi-processor efficiency, so that her kn/s figures provides a true speedup. Other engines typically display literal node counts, which include many "wasted" nodes, and this can give a false impression of a better multi-processor scaling."

[Rolf]

look here, I guess you can all read without membership.

http://rybkaforum.net/cgi-bin/rybkaforu ... ?pid=87852

Here you will learn how the KN/sec is done and meant. Also about the differences between Rybka 2 and 3. Hope this helps.

Could you please take a look, Christophe?

It seems to me that you dont even understand the point of the argument. The following quote is the only thing that I saw in the readme that's somewhat relevant, but that only explains the SMP scaling..From looking at Vas's second reply, it looks to me like he is just being stubborn and he is rather evasive in answering a direct question.

(*) Rybka's kn/s figure doesn't increase much when she runs on machines with more processors, compared to other programs. Does this mean that she scales worse on such hardware?

No. Rybka's kn/s on multi-processor machines reflects the estimated multi-processor efficiency, so that her kn/s figures provides a true speedup. Other engines typically display literal node counts, which include many "wasted" nodes, and this can give a false impression of a better multi-processor scaling."

Dont worry about my reading. Look at this, is it reveiling for you or not?

Nodes and depths are just internal counters, they don't have any user meaning or tell you anything about the engine. You can only compare Rybka 3 nodes with Rybka 3 nodes and Rybka 3 plies (ie. depths) with Rybka 3 plies.

I'll add this to the FAQ.

Vas


Why did you choose to have nodes and depths without user meaning (as opposed to other engines which do, I think, have user meaning)? I think users would generally prefer to have meaningful info.

Of course the important thing is the PV output, in which Rybka obviously excels. But I am curious!

Thanks


Sure, my statement applies to all programs.

Vas

+++++++++++++++++


So, could it be that people simply dont understand what Vas is really telling them? For me this is a very clear statement and answering. It is in complete agreement with what I read from Bob about KN count and what it meant in (ALL) programs or entities.

[Tony Thomas]

Dont worry about my reading. Look at this, is it reveiling for you or not?

Nodes and depths are just internal counters, they don't have any user meaning or tell you anything about the engine. You can only compare Rybka 3 nodes with Rybka 3 nodes and Rybka 3 plies (ie. depths) with Rybka 3 plies.

I'll add this to the FAQ.

Vas


Why did you choose to have nodes and depths without user meaning (as opposed to other engines which do, I think, have user meaning)? I think users would generally prefer to have meaningful info.

Of course the important thing is the PV output, in which Rybka obviously excels. But I am curious!

Thanks


Sure, my statement applies to all programs.

Vas


+++++++++++++++++


So, could it be that people simply dont understand what Vaas is really telling them? For me this is a very clear statement and answering. It is in complete agreement with what I read from Bob about KN count and what it meant in (ALL) programs or entities.

That still doesnt explain why he divides the node count by 10.

[Rolf]

Dont worry about my reading. Look at this, is it reveiling for you or not?

Nodes and depths are just internal counters, they don't have any user meaning or tell you anything about the engine. You can only compare Rybka 3 nodes with Rybka 3 nodes and Rybka 3 plies (ie. depths) with Rybka 3 plies.

I'll add this to the FAQ.

Vas


Why did you choose to have nodes and depths without user meaning (as opposed to other engines which do, I think, have user meaning)? I think users would generally prefer to have meaningful info.

Of course the important thing is the PV output, in which Rybka obviously excels. But I am curious!

Thanks


Sure, my statement applies to all programs.

Vas


+++++++++++++++++


So, could it be that people simply dont understand what Vaas is really telling them? For me this is a very clear statement and answering. It is in complete agreement with what I read from Bob about KN count and what it meant in (ALL) programs or entities.

That still doesnt explain why he divides the node count by 10.

What tells you tht there must be an answer at all? For me the quote above explains your difficulty but you must think a bit for yourself. The answer is it doesnt matter, the meaning is always only refering to the one particular program just like all others have their own and particular counting. You can ask for consitance only for a single program at once. The rest is speculation. Interesting that I can understand that quite easily because this isnt anything of computerchess programming internals. Could you as higher expert in the field explain to me what your and others problem really is?

[Graham Banks]

That still doesn't explain why he divides the node count by 10.

From my point of view as somebody who enjoys playing engines against each other, being able to see the pv is much more interesting.
The node count means nothing to me.

Although we all have our preferences as to what we'd like to see, the bottom line is that the engine authors are free to show whatever they like. There are no industry rules that they must adhere to in that regard.

Does anybody accuse Lance of trying to hide something because he chooses not to display Thinker's PV? There are of course a few other engines that don't produce this information also.

There's an interesting thread in the Rybka forum that seems to indicate a deal of hypocracy.
http://rybkaforum.net/cgi-bin/rybkaforu ... l?tid=6210

This clone accusation complex seems to be an all consuming affliction for some these days. Disappointing really, because it detracts from the good and useful work that they do produce.

[ernest]

bla bla bla...

It seems to me that you dont even understand the point of the argument.

My God, Tony, you should know better than starting an argument with HIM!

[Rolf]

bla bla bla...

It seems to me that you dont even understand the point of the argument.

My God, Tony, you should know better than starting an argument with HIM!

was that ad hominem?

[bob]

look here, I guess you can all read without membership.

http://rybkaforum.net/cgi-bin/rybkaforu ... ?pid=87852

Here you will learn how the KN/sec is done and meant. Also about the differences between Rybka 2 and 3. Hope this helps.

Could you please take a look, Christophe?

That is all well and good. But it doesn't explain Rybka 1 and 2 giving false numbers, does it? And they were where the problem was first observed and discussed...

One can say anything one wants, but in this case, unfortunately, the facts are on the other side of the opinion...

[bob]

look here, I guess you can all read without membership.

http://rybkaforum.net/cgi-bin/rybkaforu ... ?pid=87852

Here you will learn how the KN/sec is done and meant. Also about the differences between Rybka 2 and 3. Hope this helps.

Could you please take a look, Christophe?

It seems to me that you dont even understand the point of the argument. The following quote is the only thing that I saw in the readme that's somewhat relevant, but that only explains the SMP scaling..From looking at Vas's second reply, it looks to me like he is just being stubborn and he is rather evasive in answering a direct question.

(*) Rybka's kn/s figure doesn't increase much when she runs on machines with more processors, compared to other programs. Does this mean that she scales worse on such hardware?

No. Rybka's kn/s on multi-processor machines reflects the estimated multi-processor efficiency, so that her kn/s figures provides a true speedup. Other engines typically display literal node counts, which include many "wasted" nodes, and this can give a false impression of a better multi-processor scaling."

For that, I can absolutely state that is is just a bunch of nonsense. The only way you can measure this "search overhead" is to run the search twice, once using one CPU and once using 4. Otherwise you can only do as I have tried to do when discussing speedups and use an "estimate" that can be wildly wrong in different positions.

This just can't be done with any degree of accuracy, and is nonsense. As most of us realize.