| |
Main
Date: 20 Feb 2008 23:38:38
From: Sanny
Subject: Why Rybka Won GetClub An Analysis.
|
I found that When Playing with Rybka it was processing the Moves on a
fast Server.
A server is 15-20 times faster than a Desktop where Applet is loaded.
I got below Information from a Website
"There are certain workstation models on the ket with a total of
eight cores today. With each core able to perform 5.2 gigaflops (5.2
billion floating-point operations per second) as a measure of peak
performance, a single workstation could perform 20.8 gigaflops. "
While a desktop works on arround 1 gigaflop. So Rybka is getting 20
times more processing Power.
Other thing is that It is written in Native Language (Assembly / C)
Which again works 10-20 times faster than an Java Applet.
So Rybka is getting Atleast 15* 10 = 150 times to 20*20 = 400 times
more processing Power.
So Roughly Rybka is getting 200-400 times more processing than
GetClub.
I found when Playing with Normal Level, Normal took 16 times more time
than Rybka So we can say Rybka was processing 400/16 = 25 times more
processing.
With 25 Times more Processing It can easily beat the Normal Level.
Incase Rybka is Run on a single Core Ordinary Desktop and Given 25
times less time than Normal Level then that will be a equal Match.
Say Normal Thinking in 1-2 min and Rybka on Ordinary Single Core
Machine at 3-4 seconds / Move. Then Both will be given equal
Processing.
I am confident Normal Level will give good Challenge to Rybka thinking
3-4 seconds on Single Core Machine.
Does anyone have Rybka on Single Core Machine? Then let me know the
Games played between Normal Level and Rybka at 3-4 Seconds.
Rybka was winning the Matches because of Fast Servers and Native Code.
In future when all will be satisfied with GetClub Chess. I will bring
a Downloadable version of GetClub Game which will be 20 times faster
than the Applet you are currently Playing.
But before I invest Money in designing that version I want to be
assured that GetClub Program is not playing any wrong move. If you
find any mistake in its game do inform me.
Bye
Sanny
Play Chess at: http://www.GetClub.com/Chess.html
|
|
| |
Date: 21 Feb 2008 01:38:10
From: help bot
Subject: Re: Why Rybka Won GetClub An Analysis.
|
On Feb 21, 3:23 am, Sanny <[email protected] > wrote:
> I only want that both get equal Processing Power. So that we can
> Compare their Evaluation Strength.
I think there is no comparison; the best we
can hope to do is to determine how much
*extra time* will compensate (if any will) for
Rybka's superior evaluation function.
In replaying many of the games between
Rybka and its commercial rivals, I noted
that there was a big difference in terms of
"understanding", not speed. The sole
exception was Zap Chess or Zappa, as
also indicated by the overall scores. So
it should come as no surprise that Rybka
can also out-understand the GC program,
not merely brute-force its way to wins
using greater processor speed (Fritz
suffices for that).
> Otherwise we are comparing Apples
> to Oranges. I am not so sure that Rybka can beat GetClub after giving
> equal footage. But we cant say anything unless the real Experiment is
> done by someone.
Well, we already know that Rybka is one
of the two strongest chess programs to ever
be developed -- not counting Deeper Blue.
And we also know that the players who
use some random chess program to play at
GetClub have been winning their games; a
few have managed perfect scores, after
contesting a score of games or more.
> > > Does anyone have Rybka on Single Core Machine? Then let me know the
> > > Games played between Normal Level and Rybka at 3-4 Seconds.
>
> > Please, give GC a fighting chance-- make that
> > the Master level. Use two separate machines,
> > so that neither program is slowed by the other.
>
> Incase you use Master Level It will just think 4 times more than
> Normal Level, So It will be able to compete with Rybka making moves at
> 16 secs on Single Core Desktop.
Here is what the old Fritz 5.32 can do at
16 seconds, from the starting position with
no openings book: the display shows an
average of 1,400,000 nodes per second,
which comes to approximately:
16 X 1400000 = 22+ million nodes examined
in just 16 seconds!
> > The single most glaring problem right now
> > is the King hiding in a corner in the endgame.
> > This alone -- if known to the opponents -- can
> > result in a massive drop in performance, in
> > ratings. It has been a deciding factor in
> > several of my games, for instance; I 'know" I
> > can win a drawn ending because of this, so
> > the pressure to succeed earlier is greatly
> > reduced.
>
> Yes King hiding problem needs to be seen But recently I found in a few
> games the King came out to fight. I found when playing with Rybka
> GetClub took the King out once. Still end game needs to be improved a
> lot. But currently I am focussing only on Middle Game. As I believe if
> you cannot win in Middle Game there is no point in teaching end game.
If the opening AND middle game are improved
to a certain point, the vast majority of human
opponents will never last to an endgame. Still,
there will always be those players who are
using another chess program, and thus, the
issue will not evaporate into thin air.
-- help bot
|
| |
Date: 21 Feb 2008 00:23:26
From: Sanny
Subject: Re: Why Rybka Won GetClub An Analysis.
|
> > In case Rybka is Run on a single Core Ordinary Desktop and Given 25
> > times less time than Normal Level then that will be a equal Match.
>
> > Say Normal Thinking in 1-2 min and Rybka on Ordinary Single Core
> > Machine at 3-4 seconds / Move. Then Both will be given equal
> > Processing.
>
> =A0 Okay, assuming the math were correct,
> Rybka would beat the GC program like
> carrots; this is because of its far superior
> evaluation function.
That is what I need to test As Processing Power is an External Matter
and Can be increased Once we know that It is evaluating Correctly.
> > I am confident Normal Level will give good Challenge to Rybka thinking
> > 3-4 seconds on Single Core Machine.
>
> =A0 No doubt you can find some way to make
> Rybka play poorly; but in a fair test, my
> guess is tha Rybka will beat GC like carrots;
> chop it like cabbage; defoliate it like agent
> orange; raze it like napalm; decimate it like
> a neutron bomb.
I only want that both get equal Processing Power. So that we can
Compare their Evaluation Strength. Otherwise we are comparing Apples
to Oranges. I am not so sure that Rybka can beat GetClub after giving
equal footage. But we cant say anything unless the real Experiment is
done by someone.
> > Does anyone have Rybka on Single Core Machine? Then let me know the
> > Games played between Normal Level and Rybka at 3-4 Seconds.
>
> =A0 Please, give GC a fighting chance-- make that
> the Master level. =A0Use two separate machines,
> so that neither program is slowed by the other.
Incase you use Master Level It will just think 4 times more than
Normal Level, So It will be able to compete with Rybka making moves at
16 secs on Single Core Desktop.
> =A0 The single most glaring problem right now
> is the King hiding in a corner in the endgame.
> This alone -- if known to the opponents -- can
> result in a massive drop in performance, in
> ratings. =A0It has been a deciding factor in
> several of my games, for instance; I 'know" I
> can win a drawn ending because of this, so
> the pressure to succeed earlier is greatly
> reduced.
Yes King hiding problem needs to be seen But recently I found in a few
games the King came out to fight. I found when playing with Rybka
GetClub took the King out once. Still end game needs to be improved a
lot. But currently I am focussing only on Middle Game. As I believe if
you cannot win in Middle Game there is no point in teaching end game.
Bye
Sanny
Play Chess at: http://www.GetClub.com/Chess.html
|
| |
Date: 21 Feb 2008 00:02:41
From: help bot
Subject: Re: Why Rybka Won GetClub An Analysis.
|
On Feb 21, 2:38 am, Sanny <[email protected] > wrote:
> So Rybka is getting Atleast 15* 10 = 150 times to 20*20 = 400 times
> more processing Power.
>
> So Roughly Rybka is getting 200-400 times more processing than
> GetClub.
That sounds about right. But the move-times
indicated in the games posted here mainly
reflect the slowness of the human operator,
since it was admitted that *all* of that time was
reported as Rybka's thinking time. In sum, we
the readers have no idea how much time was
actually spent by Rybka to ponder its moves.
> I found when Playing with Normal Level, Normal took 16 times more time
> than Rybka So we can say Rybka was processing 400/16 = 25 times more
> processing.
>
> With 25 Times more Processing It can easily beat the Normal Level.
Generally, they go by "nodes". If Rybka
indicates that it has examined roughly a
trillion nodes, then you would need to give
the GetClub applet sufficient time to do the
same (around six years, I expect).
> In case Rybka is Run on a single Core Ordinary Desktop and Given 25
> times less time than Normal Level then that will be a equal Match.
>
> Say Normal Thinking in 1-2 min and Rybka on Ordinary Single Core
> Machine at 3-4 seconds / Move. Then Both will be given equal
> Processing.
Okay, assuming the math were correct,
Rybka would beat the GC program like
carrots; this is because of its far superior
evaluation function.
> I am confident Normal Level will give good Challenge to Rybka thinking
> 3-4 seconds on Single Core Machine.
No doubt you can find some way to make
Rybka play poorly; but in a fair test, my
guess is tha Rybka will beat GC like carrots;
chop it like cabbage; defoliate it like agent
orange; raze it like napalm; decimate it like
a neutron bomb.
> Does anyone have Rybka on Single Core Machine? Then let me know the
> Games played between Normal Level and Rybka at 3-4 Seconds.
Please, give GC a fighting chance-- make that
the Master level. Use two separate machines,
so that neither program is slowed by the other.
> But before I invest Money in designing that version I want to be
> assured that GetClub Program is not playing any wrong move. If you
> find any mistake in its game do inform me.
The single most glaring problem right now
is the King hiding in a corner in the endgame.
This alone -- if known to the opponents -- can
result in a massive drop in performance, in
ratings. It has been a deciding factor in
several of my games, for instance; I 'know" I
can win a drawn ending because of this, so
the pressure to succeed earlier is greatly
reduced.
-- help bot
|
|
