I think that all the leading programmers learned things about Rybka either from the Strelka release, the fraudulent release of data, or by other perfectly legitimate methods, and that many of the ideas were incorporated in various new programs. I only talked about Shredder 11 because that was the only post-strelka program I had tested then.
![[de]](/mwf/v12/flags/de.png "Germany")
... used many ideas from Rybka to get the nearly one hundred Elo jump over Shredder 10.
He Larry, you forgot some engines. F11 made a 100 Elo jump over F10 and DeepSjeng 3.0 made a 100 Elo jump over 2.7.
I think you have to correct your statement. It sounds like he does illegitimate (illegal?) things. We are not interested in rumors or hearsay when you accuse SMK. Facts or excuse!
PS: SMK would be the last I suspect.
He Larry, you forgot some engines. F11 made a 100 Elo jump over F10 and DeepSjeng 3.0 made a 100 Elo jump over 2.7.
I think you have to correct your statement. It sounds like he does illegitimate (illegal?) things. We are not interested in rumors or hearsay when you accuse SMK. Facts or excuse!
PS: SMK would be the last I suspect.
![[za]](/mwf/v12/flags/za.png "South Africa")
PS: SMK would be the last I suspect.True Roland, I don't think SMK would do such a thing. I hope sends the post again.
![[ch]](/mwf/v12/flags/ch.png "Switzerland")
+1
Hi Filipov,
Now! You have heard the term " Every Man has his price" this I am sure. That is one example of " amnesia" . When a man sells himself out short. SMK would never do that- it just would not follow- (but every man has his price?!). Another is the more command where by one falls into the " Throws of confusion " due to extenuating circumstances so much so where he forgets himself. Forgets who he is and then loses his " Center" (self) that is in my book, amnesia. So you see- you don't have to buy it - it's free!!!
sidserious-my elevator is now going to the top floor. Oh, Boy!!!
Now! You have heard the term " Every Man has his price" this I am sure. That is one example of " amnesia" . When a man sells himself out short. SMK would never do that- it just would not follow- (but every man has his price?!). Another is the more command where by one falls into the " Throws of confusion " due to extenuating circumstances so much so where he forgets himself. Forgets who he is and then loses his " Center" (self) that is in my book, amnesia. So you see- you don't have to buy it - it's free!!!
sidserious-my elevator is now going to the top floor. Oh, Boy!!!
As noted elsewhere, I meant no such thing; only that most of the latest versions of top programs seem to play more like Rybka than the previous generation, according to many posts on this forum at least.
CCRL statistics suggest this, look at Pond hit and Eval diff, the difference f10/f11 is clear
http://computerchess.org.uk/ccrl/4040/cgi/engine_details.cgi?match_length=30&print=Details&eng=Fritz%2010#Fritz_10
http://computerchess.org.uk/ccrl/4040/cgi/engine_details.cgi?match_length=30&print=Details&eng=Fritz%2011#Fritz_11
Another example is Naum.
But Shredder 11 seems a "distinct entity", at least from CCRL statistics point of view
http://computerchess.org.uk/ccrl/4040/cgi/engine_details.cgi?match_length=30&print=Details&eng=Deep%20Shredder%2011%2064-bit%204CPU#Deep_Shredder_11_64-bit_4CPU
http://computerchess.org.uk/ccrl/4040/cgi/engine_details.cgi?match_length=30&print=Details&eng=Fritz%2010#Fritz_10
http://computerchess.org.uk/ccrl/4040/cgi/engine_details.cgi?match_length=30&print=Details&eng=Fritz%2011#Fritz_11
Another example is Naum.
But Shredder 11 seems a "distinct entity", at least from CCRL statistics point of view
http://computerchess.org.uk/ccrl/4040/cgi/engine_details.cgi?match_length=30&print=Details&eng=Deep%20Shredder%2011%2064-bit%204CPU#Deep_Shredder_11_64-bit_4CPU
![[fi]](/mwf/v12/flags/fi.png "Finland")
Use url tags, please!
Some CCRL URLs are very long... I use this in my favourites, to create a Tiny URL for a page immediatly:
javascript:void(location.href='http://tinyurl.com/create.php?url='+location.href)
javascript:void(location.href='http://tinyurl.com/create.php?url='+location.href)
![[nl]](/mwf/v12/flags/nl.png "Netherlands")
Very good observing, thanks for pointing this out!
> CCRL statistics suggest this, look at Pond hit and Eval diff, the difference f10/f11 is clear
>
> http://computerchess.org.uk/ccrl/4040/cgi/engine_details.cgi?match_length=30&print=Details&eng=Fritz%2010#Fritz_10
>
> http://computerchess.org.uk/ccrl/4040/cgi/engine_details.cgi?match_length=30&print=Details&eng=Fritz%2011#Fritz_11
>
> Another example is Naum.
>
> But Shredder 11 seems a "distinct entity", at least from CCRL statistics point of view
>
> http://computerchess.org.uk/ccrl/4040/cgi/engine_details.cgi?match_length=30&print=Details&eng=Deep%20Shredder%2011%2064-bit%204CPU#Deep_Shredder_11_64-bit_4CPU
It seems too that you are part of Shredder team or a supporter ;)
A note on CCRL ponder hits statistics: The Fritz interface does not give a note if there WAS a ponder hit, it gives a note if a reply was NOT expected by the other engine (by adding the expected move as a comment). In other words, the Fritz GUI provides information about ponder non-hits, or ponder disagreements.
I think that is why we find engines like Thinker 5.1 or Micro-Max in the list of "Most similar pairs". These engines do not send a pv, or not in a way Fritz can read it, so the GUI will never find a ponder disagreement for these engines. Therefore, for example the 85.2% for Thinker with Fritz 8 can only mean that Fritz 8 caused 14.8% ponder disagreements in their games, but we do not know how often Thinker disagreed. With another ~20%, it would result in a more or less normal total p.h.%.
(With "normal" engines it is a mix of ponderhits from both.)
http://tinyurl.com/585wjd
P.S. I found Shredder only once, in these correlation lists: Deep Shredder 9 is "most similar" to Shredder 9 :-) (by eval difference). Surprisingly, they played 399 test games against each other.
I think that is why we find engines like Thinker 5.1 or Micro-Max in the list of "Most similar pairs". These engines do not send a pv, or not in a way Fritz can read it, so the GUI will never find a ponder disagreement for these engines. Therefore, for example the 85.2% for Thinker with Fritz 8 can only mean that Fritz 8 caused 14.8% ponder disagreements in their games, but we do not know how often Thinker disagreed. With another ~20%, it would result in a more or less normal total p.h.%.
(With "normal" engines it is a mix of ponderhits from both.)
http://tinyurl.com/585wjd
P.S. I found Shredder only once, in these correlation lists: Deep Shredder 9 is "most similar" to Shredder 9 :-) (by eval difference). Surprisingly, they played 399 test games against each other.
Thanks for the clarification.
Strelka is considered in the same family than Rybka, so it does not appear in that list
Strelka is considered a clone of Rybka 1 beta and the statistics are 71.4% and 0.23
Strelka is considered in the same family than Rybka, so it does not appear in that list
Strelka is considered a clone of Rybka 1 beta and the statistics are 71.4% and 0.23
![[mx]](/mwf/v12/flags/mx.png "Mexico")
> only that most of the latest versions of top programs seem to play more like Rybka than the previous generation, according to many posts on this forum at least.
Yes, Shredder 10 was one of the most aggressive engines of its time (not counting King Attacks) and the change in playing style to Shredder 11 is dramatic. You just have to watch the games to realize that something happened and it doesn't feel much like Shredder anymore. (I apologize in advance if this sounds accusatory, that's not my point, but a different approach seems to have taken place.)
Eventually I'll solve this mystery.
I think the gain of ~70 Elo in the Fritz match is quite normal (~35 Elo per doubling cores). R2 vs R3 is special. Shredder is the favorite opponent of Rybka (look to CEGT and CCRL).
My suggestions:
1. When R3 sp plays a match take a longer tc than 1+1 (e.g. 4+1 or even 5+1). TC 1+1 is to short for sp!
2. Take more opponents to check what is right (+130 or +190 can´t be right).
I think the gain of ~70 Elo in the Fritz match is quite normal (~35 Elo per doubling cores). R2 vs R3 is special. Shredder is the favorite opponent of Rybka (look to CEGT and CCRL).
My suggestions:
1. When R3 sp plays a match take a longer tc than 1+1 (e.g. 4+1 or even 5+1). TC 1+1 is to short for sp!
2. Take more opponents to check what is right (+130 or +190 can´t be right).
If 1'+1" is too short for sp, why is it only too short against another Rybka, but not against the other engines? As for Shredder, first of all my impreession was the opposite of yours, but anyway it doesn't matter as the issue is not which opponent scores well or poorly but which one shows a bigger rating differential when Rybka goes from 1 core to 4 cores, an entirely different question. Taking other opponents will tell us what is the normal effect of going from 1 to 4 cores, but I want to know why the numbers are so drastically different in these two cases, especially the Rybka vs. Rybka case.
If 1'+1" is too short for sp, why is it only too short against another Rybka, but not against the other engines?
Where are the data? I can´t see that it isn´t too short against other engines. It´s too short against Shredder too. When we only have data about 3 engines (F, S and R2) we can´t say exactly what´s normal, only guess. Maybe Elo differences in the Shredder and R2 cases are not drastically different from other matches in 1+1 tc? We don´t know.
Where are the data? I can´t see that it isn´t too short against other engines. It´s too short against Shredder too. When we only have data about 3 engines (F, S and R2) we can´t say exactly what´s normal, only guess. Maybe Elo differences in the Shredder and R2 cases are not drastically different from other matches in 1+1 tc? We don´t know.
I don't have other strong engines to test against right now, though I may get others. But the difference between the result against Fritz and against Rybka 2.32a is too huge to be explained by normal ideas; I'm testing a theory about this now.
I'm testing a theory about this now.
Please let me know if you was succsessful.
My theory: Good pruning, one of Rybkas strength, needs some time to be successful (more than 1+1 on sp). What´s average ply depth from R3 in 1+1 games on sp?
Please let me know if you was succsessful.
My theory: Good pruning, one of Rybkas strength, needs some time to be successful (more than 1+1 on sp). What´s average ply depth from R3 in 1+1 games on sp?
It's true that average depth at this speed is low on sp; I don't watch carefully this number, but maybe somewhere around 10 plies. But again, if that's such a problem, it should hurt against Fritz as well as against Rybka 2.3.2a.
My first theory about this was not confirmed by tests. But meanwhile I got a new version today, and finally this one on one core is narrowly beating Rybka 2.3.2a on 4 cores after 166 games at 1'+1" (+14 Elo). I don't think it could be that dramatically better than the last couple versions, but this result is a big improvement. Anyway it looks like we have answered your wish for a Rybka 3 that could give Rybka 2.3.2a 4 cores to 1 handicap straight up, and at the 1'+1" time limit you deem too fast for 1 core!
Are you basically saying, Rybka 3 on one core beats Rybka 2 on 4 cores?!
Since 2005, we were witnessing a whole series of surprises. - I am flabbergasted. :-)
Since 2005, we were witnessing a whole series of surprises. - I am flabbergasted. :-)
Yes, that's exactly what I'm saying. Final result +45=92-36 for +18 Elo. I'm rather amazed myself.
Great! And you will see at Blitz (not bullet!) it will be +30. :-)
![[cr]](/mwf/v12/flags/cr.png "Costa Rica")
If Rybka 3 SP finally be able to defeat Rybka 2.3.2a on 4 cores, I think this new version will have overcome thoroughly with all the expectations. This overcomes everything. An improvement of this class It is almost worthy of a prize Nobel :-)
Really thanks to you and Vas by this amazing achievement. It's quite difficult to believe a result like that Larry, it's really impressive.
Do you believe that Rybka 3 SP can deafeat 2.3.2a 4 cores but under slower time controls? I think it could be good to re-run some mini match -less games- but under slower time controls. Why not with 1+3 or 2+2 time controls?
Regards,
Gaмßito.
Really thanks to you and Vas by this amazing achievement. It's quite difficult to believe a result like that Larry, it's really impressive.
Do you believe that Rybka 3 SP can deafeat 2.3.2a 4 cores but under slower time controls? I think it could be good to re-run some mini match -less games- but under slower time controls. Why not with 1+3 or 2+2 time controls?
Regards,
Gaмßito.
Rerunning at a slightly slower time control is pointless; the results would be very similar (except for sample error), and I don't have the CPU time to rerun at a much slower time control for the needed hundreds of games.
Having a machine that is 3 times as fast you will get what ELO-win approximately?
I wouldn't assume that this would be more than 100 ELO points.
Rybka3 ist estimated more than 100 ponts stronger than Rybka2.
(big Respect, if this is true!!!)
But why is it astonishing then, that the one-core-Rybka3 wins against the 4-core-Rybka2?
Quap
I wouldn't assume that this would be more than 100 ELO points.
Rybka3 ist estimated more than 100 ponts stronger than Rybka2.
(big Respect, if this is true!!!)
But why is it astonishing then, that the one-core-Rybka3 wins against the 4-core-Rybka2?
Quap
Only because of my experience that playing related programs seems to magnify the effect of a hardware handicap.
Das wären maximal 24 Elopunkte, klingt fast wie statistisches Rauschen.
I think you missed the point here...
WOW! is an understatement. We need this exact Rybka 3 beta version! Don't change anything, it's that what we want. We don't want something to be broken by changing things. :-)
But i wonder if it's because of MUCH better CPU use efficiency(speedup) of Rybka 3 over Rybka 2.3.2a.
So it would be interesting if you can match this latest Rybka 3 against Rybka 2.3.2a with a 1:4 time handicap.
That would, comparing with the result on 1:4 cores you gave, show interesting things as how much speedup efficiency has improved....
If for example the 1:4 time handicap give +25 ELO for Rybka 2.3.2a then speedup efficiency has been improved a lot.
If the 1:4 time handicap give around +18 ELO for Rybka 3 then speedup efficiency hasn't been improved by much, but general playing ability of Rybka 3 has been improved much more than the expected.
If the 1:4 time handicap give +50 ELO for Rybka 3 then speedup efficiency would become worse. :-) That case it would mean general playing ability of Rybka 3 has amazingly drastically improved much more than the expected. It would also mean another thing: That something is broken in Rybka's multiprocessor skills, so if that speedup bug is fixed then Rybka would have another +20-30 ELO.
But i wonder if it's because of MUCH better CPU use efficiency(speedup) of Rybka 3 over Rybka 2.3.2a.
So it would be interesting if you can match this latest Rybka 3 against Rybka 2.3.2a with a 1:4 time handicap.
That would, comparing with the result on 1:4 cores you gave, show interesting things as how much speedup efficiency has improved....
If for example the 1:4 time handicap give +25 ELO for Rybka 2.3.2a then speedup efficiency has been improved a lot.
If the 1:4 time handicap give around +18 ELO for Rybka 3 then speedup efficiency hasn't been improved by much, but general playing ability of Rybka 3 has been improved much more than the expected.
If the 1:4 time handicap give +50 ELO for Rybka 3 then speedup efficiency would become worse. :-) That case it would mean general playing ability of Rybka 3 has amazingly drastically improved much more than the expected. It would also mean another thing: That something is broken in Rybka's multiprocessor skills, so if that speedup bug is fixed then Rybka would have another +20-30 ELO.
I don't understand your logic. Whatever MP speedup has been achieved in Rybka 3 has nothing to do with this test, because Rybka 3 was running on 1 core. We already know the apeedup in Rybka 2.3.2 from 1 to 4 cores (somewhere in the 2.6-2.8 range, I forget exactly). So this was in effect a 2.7 -1 time handicap, roughly.
Yes you're right. For some reason i took this in my head as 4 cores Vs 8 cores.
![[be]](/mwf/v12/flags/be.png "Belgium")
What changes have been done ?
Improved search ? longer thinking time for R3 ? other things ?
Improved search ? longer thinking time for R3 ? other things ?
Since last week's version, there were some small eval changes, probably worth no more than 1 Elo, and some search changes, which appear to have helped.
You know! There seems to be this gentle but persuasive Rybka generational senility that sets in with regards to versions. One can only tell folktales about the way Rybka 2.1 used to play chess and the mighty coming of Rybka 2.2n2 w32. Oh, my God! What a magnificent version. Then came the others more versions and...then, well then, Larry came on to the scene and there was the intro of Rybka2.3.lk.(mp).w32. Now, we are left with this...post mortems and hopeful thoughts.
![[us]](/mwf/v12/flags/us.png "United States")
Robert, do you know whether persistent hash is in the latest version and, if not, whether it will be in the final version?
Haha, I know that you're not part of the Rybka team; I just figured I'd ask anyway. :-P
Haha, I know that you're not part of the Rybka team; I just figured I'd ask anyway. :-P
I really couldn't say, but I'll bet Harvey knows. Now! Whether he is willing to turn over the projects secrets is another story.
You, know that we all have to meet in the cyber chasm where there is no bloody light to speak off and know one gives you a straight answer. So, what the hell do I know anyhow? (Besides, there are members of the team all around their names remain secret-they know who I am).
I am part of the Aquarium Team! Dude
What good is a fish without a bowl? And, if you remember...I was the idiot that was loud mouthing off right from the outset, " PUT THE FISH IN A BOWL!" No one listened. I went to Convekta and convinced them to talk to the forum to see if the product would be there to sell. They came- found a market and maded the connection. The rest is history, Jack!
NOW! " THE FISH IS IN THE BOWL " .................................................................................hot cha cha
sidserious- " David "- this one has got to be for you- it from the beginning!!!
However, will the fish-like in the PBS commercial dive out of the bowl-right out a window and on to an umbrella, sliding off into a puddle, only to flop out of the way of an on coming car just in the nick of time, then to,...flip into an empty water receptacle waiting to go back to be refilled. Then, when cross and expansion bridge, that mighty little fish jumps out of the jug into the rushing waters below- and, swims, with the rest of the cyber free, chess engines. Like it was meant to be, right from the very beginning.
NOW! " THE FISH IS IN THE BOWL " .................................................................................hot cha cha
sidserious- " David "- this one has got to be for you- it from the beginning!!!
However, will the fish-like in the PBS commercial dive out of the bowl-right out a window and on to an umbrella, sliding off into a puddle, only to flop out of the way of an on coming car just in the nick of time, then to,...flip into an empty water receptacle waiting to go back to be refilled. Then, when cross and expansion bridge, that mighty little fish jumps out of the jug into the rushing waters below- and, swims, with the rest of the cyber free, chess engines. Like it was meant to be, right from the very beginning.
Hi Sid,
I haven't seen your posts in a long time. Where have you been?
I haven't seen your posts in a long time. Where have you been?
Hi Filipov,
I hope all is well with you in your part of the world. I've had Amnesia!!!
I hope all is well with you in your part of the world. I've had Amnesia!!!
I am here dedicating my 5500th post to you. :-)
Rub it in!!!!
I caught, Paul's nice words for me and you, Turbo. It takes time and a great deal of consideration for each to achieve that kind of understanding and one of the biggest steps in that process, as it is a process, was the willingness to open ourselves up to each other on a number of levels. The strength of character and dignity of respecting each others boundaries allowed for us to grow as people and take shape and become real in each others lives. The amazing thing is that Paul is across the Pond, yet feels like he lives next door.
I caught, Paul's nice words for me and you, Turbo. It takes time and a great deal of consideration for each to achieve that kind of understanding and one of the biggest steps in that process, as it is a process, was the willingness to open ourselves up to each other on a number of levels. The strength of character and dignity of respecting each others boundaries allowed for us to grow as people and take shape and become real in each others lives. The amazing thing is that Paul is across the Pond, yet feels like he lives next door.
> I am here dedicating my 5500th post to you. :-)
Where's your sixth pawn??!
![[au]](/mwf/v12/flags/au.png "Australia")
> > I am here dedicating my 5500th post to you. :-)
>Where's your sixth pawn??!
If the Forum Challenge were bughouse, maybe he'd have a use for those pawns... Then again, they're probably the wrong colour. :-P
I was half expecting a lolcat.
That is perfect!
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill