"Well, it is important to keep in mind that this [Larabee] is first and foremost NOT a GPU. It's a CPU. A many-core CPU that is optimized for data-parallel processing." (excerpt from Anandtech's report on Intel's future discrete graphics card: http://anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3367)
My guess is that Monte Carlo could put to some use those Pentium-like processors, even if they're slow on their own, there will be about 32 or 64 of them to begin with (next year?).
>Could this information, somehow, change your assessment?:
> ...optimized for data-parallel processing."
No. And I wouldn't trust a website that can't even get their links working half the time (the link to the second page was unclickable)!
Surely there must be better ways to dismiss a source for information.
I gives me the impression that you just can't be bothered with the subject.
What kind of tournament you can play with R 3 under Aquarium? Rounds, simultaneously, switzer system , K.O.?
Is there the possibility: "Last game before interruption? (like under Arena)
New types of tournaments can appear after release.
since I didn't read anything about it in the last 2 month I wonder if rybka3 supports this and which GUI supports this?
I just repeat my opinion, that this is MUCH more worth than aquariums IDeA, monte carlo and jeroens book together.
The function "persistent hash" it is also useful in the game blitz?
Doesn't It slow down the engine?
The other options of formulation, of "persistent hash" (Play Depth,Write Depth and Size), is it worthwhile to leave it as from default, indifferently from the hardware?
Besides Rybka3 go on my PC to half speed (in K/n) of Rybka 2.3.a.
Is this normal?
The slowdown from the persistent hash should be negligible.
Rybka 3 gives roughly half the kn/s of Rybka 2.3.2a.
Can you add a little star (*) before every new question in the FAQ you add? I'm updating the FAQ on our website and it's easier to see which one is new then. When I added it to the website I'll remove the stars. And yes, I think having it on the website is better than just in the forum :)
ok, I'll start doing that now.
You might want to wait a few days, I suspect that this thing is still going to grow a bit. You can even link to the forum for now and then make it "official" later. Up to you ..
ok, sounds good. So we have three perfectionists on the Rybka team (Larry and Jeroen are the other two) :)
Seriously, as a chess player you have to be a perfectionist in some way.
> "perfectionist in some way"
Isn't that an oxymoron?
"Preserve Analysis" Saved Hash File: Do I have to write in the window where is written "<empty> " the name of folder where I want that the file of log to be written?
I have tried to write in the window "Saved Hash File : "C :\". But any files of log have not been created.
I have Shredder 8 GUI. Do I perhaps have to write not the run of the folder but the filename? Example:
> I have tried to write in the window "Saved Hash File : "C :\". But any files of log have not been created.
You need the full path, including the filename. For example, C:\SavedHash.rsh
The option Montecarlo, is only present in the version Aquarium? And in the version Chessbase?
In the version UCI Engines, are only present the engines and therefore I are not able to use this option?
It is not well clear, in the presentation online.
I will look at it more deeply, but to me it seems more one third the kn/s :-p
> I will look at it more deeply, but to me it seems more one third the kn/s :-p
To me too. With Rybka 2.3.2a it was about 42kn/s, with Rybka 3 is about 15kn/s. I was getting worried but Rybka destroys the opposition as expected so more kn/s would be overkill :-P
In endgames I have a read of tablebase + speed with Rybka 3 . Particularly, I have less job on the HD, it does' less noise, and it is faster to load the tablebase. This to make me to think that the hash tableses work better in Rybka 3.
Besides with Ram Booster 2, I see that Rybka 3, loaded + memory of Rybka 2.3.2.a. Both memory RAM both file of pagin used (this I see with Task Manager)
As for tablebases, I noticed that Vas might be fooling us into thinking he doesn't care about them very much :-): R3 set on Normaly (Nalimov usage) consults an order of magnitude TBs more often then R232a did on Normaly. I don't know when both are set to default (Rarely), but this means Vas made her look at them very much more.
Can you show an example?
I would say close to 35%!
Infinite analysis, start_position
32-bit XP Home
Core 2 Duo @3GHz
only 1 core (for reproducibility) Max CPUs=1
Engine: Rybka 3 32-bit (512 MB) by Vasik Rajlich, Larry Kaufman
12.00 0:02 +0.11 1.Nf3 Nf6 2.Nc3 Nc6 3.d4 d5 4.Bf4 Nh5 5.Bg3 (78.990) 33
13.01 0:06 +0.14 1.Nf3 Nf6 2.Nc3 d5 3.d4 e6 4.g3 Nc6 5.Bg2 Bd6 (210.522) 34
14.01 0:12 +0.09 1.Nf3 Nf6 2.Nc3 d5 3.d4 e6 4.e3 Nc6 5.Bd3 Be7 6.O-O (418.712) 33
14.12 0:28 +0.13 1.e4 Nc6 2.Nf3 e5 3.Bb5 Nf6 4.O-O Bc5 5.c3 O-O 6.d4 (948.929) 33
15.01 0:44 +0.19 1.e4 Nc6 2.Nf3 Nf6 3.e5 Nd5 4.Nc3 Nxc3 5.dxc3 (1.415.555) 32
Engine: Rybka 2.3.2a mp 32-bit (512 MB) by Vasik Rajlich
12.03 0:02 +0.12 1.d4 d5 2.Bf4 Nc6 3.Nc3 Bf5 4.Nf3 Nf6 5.Nh4 (225.415) 105
13.01 0:06 +0.12 1.d4 d5 2.Nf3 Nc6 3.Bf4 Bf5 4.Nc3 e6 5.e3 Nf6 (690.678) 113
14.01 0:08 +0.12 1.d4 d5 2.Nf3 Nc6 3.Bf4 Nf6 4.Nc3 Nh5 5.Qd2 Nxf4 6.Qxf4 e6 (970.386) 111
15.01 0:15 +0.13 1.d4 d5 2.Nf3 Nc6 3.Bf4 Nf6 4.Nc3 Nh5 5.Qd2 Nxf4 6.Qxf4 e6 7.Nb5 (1.604.021) 109
16.01 0:26 +0.10 1.d4 d5 2.Nf3 Nc6 3.Bf4 Nf6 4.Nc3 Nh5 5.Be3 g6 6.g3 Bg7 7.Bg2 (2.976.265) 115
Ratio: 33/110 = 30%
Can you post your result?
You see, Rybka 3 go to half speed
> You see, Rybka 3 go to half speed
Thank you, Albitex.
True, with the position you indicated, and infinite analysis (512MB hash) I get around 42 to 45% kn/s for Rybka 3, quite a bit more than my 30-35% for the start_position.
If you have a moment, can you try infinite analysis of the start_ position, same hash, 1 core only (for reproducibility).
>perfectionists aren't perfect the first time, they're perfect the last time.
Vas... this is of no comfort to the perfectionist. :-P
>Rybka 3 gives roughly half the kn/s of Rybka 2.3.2a.
What I can't understand is... why there isn't some sort of slider or box (like contempt) labeled KPF for Rybka 3 (kN/s peace-o-mind factor). Default could have been... say x4. Making it adjustable would compensate e.g. the release of new hardware. Nehalem is released... am I worried? Of course not; I just bump up the KPF to a "new" default setting.
> why there isn't some sort of slider or box (like contempt) labeled KPF for Rybka 3 (kN/s peace-o-mind factor). Default could have been... say x4
This logically belongs in the GUI, not the engine :)
I was being totally facetious with Vas, because I see complaint, after complaint, after complaint... and where is the testing! Your picture shows me absolutely nothing in terms of comparing kN/s.
I have compared 2.3.2a kN (or mN) output on identical positions between my quad (amd) and octa (intel). My question: how can the same engine on the same position come up with differing amount of nodes searched at comparable depths? My answer: I could care less... because if I knew "the truth," how would that effect my use of Rybka? But what you are doing is even more radical...
You are trying to use 2 differently calibrated ways of measuring kN/s and then saying, "Aha!" (or am I wrong, Vas?) If one observed a car and measures speed with a radar gun calibrated for kph, seeing a displayed speed of 88... then a 2nd car travelling at the identical speed (but unbeknownst to you, you bumped a button that sets the calibration to mph). "Wow! What a slow poke! He is only doing 55!"
To me, Vas could have eliminated 99% of this discussion, if he had built in that magical, mystical 4x multiplier (to the current kN/s output) which varied slightly with the number of cpu cores. Or just take the kN/s output and "add" a zero to the end of it. But he is the big kahuna (doing as he sees fit)... if we are going to challenge him on something, it should be with facts / supporting data. Otherwise, we can take note that it is different, "Wow, it's different..." and then move on.
>This logically belongs in the GUI, not the engine
> You are trying to use 2 differently calibrated ways of measuring kN/s and then saying, "Aha!" (or am I wrong, Vas?) If one observed a car and measures speed with a radar gun calibrated for kph, seeing a displayed speed of 88... then a 2nd car travelling at the identical speed (but unbeknownst to you, you bumped a button that sets the calibration to mph). "Wow! What a slow poke! He is only doing 55!"
This sounds like some of the cops I've had the pleasure of meeting :)
The function Random, seems not to work in Rybka 3.
This both in Aquarium GUI, both planning the config. file for the chessbase GUI.
I have to analyze a position of endgames. If I make matches Rybka3 Random vs Rybka3 Random, or in Aquarium "random engine matchs" with Rybka3, I always have the same 1° move.
The only way, with which I have succeeded in analyzing more variants is : Rybka3 Random vs. Rybka2.3.2.a Random matchs. But also in this way, I have a problem : I have planned the config.file to "randomize=5", to have 5 principals variants . But Rybka plays all the 23 possible moves beginning from that positions.
There is something that doesn't work. ??
The position is : [FEN "2k5/ppp4p/8/8/5r2/P7/P4PKP/3R4 b - - 0 1"] . Rybka3 Random play always 1. ... c5 . Only this .
> "randomize=5", to have 5 principals variants .
centipawns = 1/100 of pawn value?
Do I have to plan this value?
, I have understood. But then, example in this case: if I plan Rybka in infinite analysis, I have to the 17/18 depth 1. .... h5. But since in Random match, I plan 10 depths for move, to 10 depths Rybka considers 1. .. h5 only as 8° / 10° values move. Then Rybka doesn't play never it (this if I use Rybka3 vs. Rybka3). While if I use Rybka3 vs. Rybka2.3.2.a (Random obviously) I get the whole 23 possible moves (excluded Rf3). Or too much, or too much little.
Besides being clearly inferior Rybka2.3.2.a in the endgames to Rybka3, the result is distorted, because Rybka3 also wins with the white (when it is draw or loss).
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill