Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Deep Rybka 5- the project continues?
1 2 Previous Next  
- - By Magnus Friedmann (***) Date 2011-08-19 05:24
This question is for Lucas: Do you know if Vas plans to release Deep Rybka 5? We want desperately this engine, I myself want Rybka to comeback improved with at least 100 ELO points. Houdini 2 will be released with a much higher strength than Houdini 1.5 a, who is stronger than Deep Rybka 4.1 so clearly, Deep Rybka 5 has to gain very much in strength
Parent - - By Silvian (***) Date 2011-08-19 05:46 Edited 2011-08-19 05:51

> Do you know if Vas plans to release Deep Rybka 5?


Not for you Mr.Thomas Freeman !:grin:

> We want desperately this engine.....


We ??????  It's about the offenders from ....mmortalchess forum ?
Desperately ?????? Robbo source code touched already his limits ? No new Ivan , Saros and other balshoi engines ?

Do you think Vasik is stupid ? NO ! Deep Rybka 5 exists on his cluster form.
But not for you and BiohAzaard team !

So sorry Thomas !
SilvianR



NR: o intrebare : de ce te iei de mana cu derbedeii aia ?
Parent - - By Magnus Friedmann (***) Date 2011-08-19 07:19
Why do you act like that? I just asked a question, that's all. I said we, because this topic has been approached a few weeks ago, as far as I remember and I know there are many chess programs addicts who are still expecting the release of Deep Rybka 5. Nu ma iau de mana, ma pasioneaza domeniul. Eventual poti sa fii mai specifica si sa vb pe email.  And the name is Friedmann.... Magnus Friedmann :)
Parent - - By Geomusic (*****) Date 2011-08-19 08:03
who likes my new avatar? It was difficult to scale down my drawing :)
Parent - - By TheHug (Bronze) Date 2011-08-19 08:03
+1 Interesting, Major dark side. What is it suppose to be. Kind of looks like the joker.
Parent - By Geomusic (*****) Date 2011-08-19 08:04
ya it's the joker.
Parent - - By Geomusic (*****) Date 2011-08-19 08:22
i changed it slightly looks a little more like him.
Parent - - By TheHug (Bronze) Date 2011-08-19 08:24
I always zoom in to get a better like at it. It looks like he has a big scar on his left eye. But it does look good. :)
Parent - - By Geomusic (*****) Date 2011-08-19 08:25
how do you zoom? It will  look like crap it u zoom it in :)
Parent - - By TheHug (Bronze) Date 2011-08-19 08:28
Do you use Fire Fox? Na it doesn't look bad I just like to try and see the detail.
Parent - - By Geomusic (*****) Date 2011-08-19 08:31
yes I use it.
Parent - - By TheHug (Bronze) Date 2011-08-19 08:33
If you have a mouse wheel click on the screen than scroll up. :smile: Then hit Ctrl+0 to return.

Edit: Sorry I made one small mistake you have to click into your mouse wheel than scroll up.
Parent - - By Uly (Gold) Date 2011-08-19 08:54
That doesn't zoom around here.
Parent - By TheHug (Bronze) Date 2011-08-19 08:57
I guess I'm special :lol:
Parent - By TheHug (Bronze) Date 2011-08-19 09:00
Here is the pic's
Attachment: Zoomout.PNG (103k)
Attachment: Zoomin.PNG (163k)
Parent - - By InspectorGadget (*****) Date 2011-08-22 19:16
It looks like that one which PortyCitySlim used to have sometime ago :lol:
Parent - By Uly (Gold) Date 2011-08-23 02:58
:smile:

Parent - - By Uri Blass (*****) Date 2011-08-20 22:48
It is a fact that I bought rybka for her playing strength.
I bought Rybka1,2,3 and I did not buy Rybka4.

If Vas has something significantly stronger than houdini for normal hardware then
I think that Vas earns less money because of his decisions not to release it.

I think that if you are clearly number 1 then it is a mistake not to release what you have and sell it(because you may not get a better opportunity in the future).
If Vas believes that the opponents need new Rybka version to get a significant improvement then I guess that he is wrong and
he may discover in the future that he is unable to catch cluster houdini even if he does not release a new rybka.
Parent - By Carl Bicknell (*****) Date 2011-08-22 19:20

> I think that if you are clearly number 1 then it is a mistake not to release what you have and sell it


if your logic is correct then Houdini is still no.1, since Vas would have thought of this.
Parent - - By Quapsel (****) Date 2011-09-02 13:40
You bought Rybka 1?
Quap
Parent - By Uri Blass (*****) Date 2011-09-02 21:03
Yes
I mean of course the first commercial version that came immediately after rybka1 beta and I got many beta versions when the last one was Rybka1.2 based on my memory.
Parent - - By Rebel (****) Date 2011-09-01 10:55

> Do you think Vasik is stupid ? NO ! Deep Rybka 5 exists on his cluster form.


I think if Vas would put some good anti-hack-disassembler stuff in R5 he could release R5 relatively safe.
Parent - - By AWRIST (****) Date 2011-09-01 11:07
Also if 20+ people would try RE on it? How much time would it take to crack the code?
Parent - By Rebel (****) Date 2011-09-01 11:34
I don't know, never underestimate the intelligence and creativity of your opponents. But I can tell from experience the moment I noticed a particular programmer was after mine I build in loads of obfuscations tricks to make hacker behaviour a hellish place to be.
Parent - - By M ANSARI (*****) Date 2011-09-03 06:52
There was some thought of putting up some USB device or some other hardware unit that would increase the difficulty of RE or hacking several fold, but somehow Vas didn't go for it.  I think with software alone you reach a point where adding stuff and obfuscating code can be counterproductive and hurt performance.  I have no doubt that some simple work on Rybka 4.1 would have at least 30 to 50 ELO increase and so maybe Vas would do that for Rybka 5.  I would bet that a threaded Rybka 4.1 with a well thought out and developed time control setup and a little more work on preventing engine stalling at depth 17+ ... and you would have a very nice ELO boost.  I still believe that Rybka 4.1 static evaluation is the most accurate of any engine out there ... although I would have to do some testing to see if it has been surpassed by Komodo 3.
Parent - - By Rebel (****) Date 2011-09-03 10:13

> There was some thought of putting up some USB device or some other hardware unit that would increase the difficulty of RE or hacking several fold, but somehow Vas didn't go for it.


All such solutions fail, hackers simply start the job after the program is loaded into memory, then dump it to hard disk.

>I think with software alone you reach a point where adding stuff and obfuscating code can be counterproductive and hurt performance.


Not at all :smile:
Parent - By Mark (****) Date 2011-09-03 13:20
My thoughts: The only solution to piracy is to keep the price very low.  The only solution to reverse engineering is frequent releases with substantial improvements.  Of course these are only partial solutions.
Parent - - By M ANSARI (*****) Date 2011-09-03 20:07
I think that is what Vas thinks as well, as he doesn't seem to have followed up on that.  It seems that he thinks the only way to keep his code safe is to have it on a secure server and thus the Rental Rybka.
Parent - - By Rebel (****) Date 2011-09-04 07:19
And might work well. If you aware of the reputation and lobbying efforts of Chessbase among GM's it's not unlikely Chessbase serves as intermediate receiving a royalty fee for every new job they deliver to Vas.

It might be the case Vas doesn't need the distribution of new versions at all.
Parent - - By M ANSARI (*****) Date 2011-09-04 07:48
Yes you might be right.  It is a shame though that he had to resort to this setup.  I would have liked to see the progression of a PC version of Rybka.  The last version of Rybka (4.1) did not feel like it was optimized for PC usage during PC beta testing.  It felt like there was a lot of potential in the engine that was not realized and it was pretty amazing how quickly the engine developed and how fixing a few bugs dramatically improved ELO.  Somehow it felt like a setup that was designed for a cluster Rybka and was forced to fit into PC clothing.  A lot of little things like engine initialization before a move, stalling sometimes at certain depth, and of course incredibly poor time management ... all made it feel like there was much more to be gained.  My feeling was that it had to do with difficulties in getting a process based engine to work as well as a threaded engine on a PC platform, and have those processes behave well or at least predictably under a Win OS.  It would be great if Rybka 5 would be built from the ground up as native PC optimized code, although with so many cores being added to new systems the differences between a PC and cluster might become a non issue.
Parent - By Rebel (****) Date 2011-09-04 07:51
And perhaps would explain a lot of Vas' decision not to stand trial.
Parent - - By Ray (****) Date 2011-09-04 08:00
Rybka was a target for the cloners and decompilers due to it's number one status for many years. None of the top engines should feel safe now. The server model is the only way to stop this, and I fully understand the route he has taken. A shame though, as you say
Parent - - By M ANSARI (*****) Date 2011-09-04 08:30
I think a lot of the "easy fruit" that was available with regards to ELO gain has already been picked.  Vas did something with Rybka 3 that dramatically increased the ELO of the engine ... emphasis on dramatic.  I remember during the beta testing of R3 where it was still very green and I put it on Playchess to test playing strength, I put it on auto accept and came back a couple of hours to see that it had added more than 200 ELO and was crushing every engine that would challenge it and had crossed the 3000 ELO mark (pretty difficult to do on Playchess especially with auto accept).  At the time it was on an Octa that was running at 3.2 Ghz, and I did have an Octa running at 4.2 Ghz which I also tested (although not on Playchess).  I immediately realized that something pretty major had happened with computer chess with regards to ELO and with regards to MP scaling.  I guess that was when Vas had just put his new search into the engine and together with the improved evaluation of Larry, it proved to be a dramatic improvement.  This easy pickings has now been accessed by everyone and so I believe progress will be back to how it used to be where an increase of 20 ELO per year was considered to be an excellent achievement.  Of course someone might come up with another revolutionary change that breaks the sound barrier again, but I really doubt it.
Parent - By Ray (****) Date 2011-09-04 08:48

> ........ and so I believe progress will be back to how it used to be where an increase of 20 ELO per year was considered to be an excellent achievement.  Of course someone might come up with another revolutionary change that breaks the sound barrier again, but I really doubt it.


I agree. I think the top engines are now at a level where extra ELO is increasingly hard to find. Of course many weaker engines can still make big jumps.

In the future big gains may come from extra cores and clustering rather then extra intelligence in the engine. Who knows what we will have on our desktops a few years from now
Parent - - By Gaмßito (****) Date 2011-08-19 09:28
Hello Magnus,

Vas already said in Nelson' video that Rybka 5 will be for the end of this year or early next year. So we can be sure that we will have a Rybka 5.

About Houdini 2 and according to Robert Houdart, it will be among september and december of this year. He didn't told me how much stronger will be, but I guess it will be at least 50 Elo points stronger than Houdini 1.5a. If this happen, then Houdini 2 will be 100 Elo points stronger than Rybka 4.1 and this is too much! So, Vas needs a very strong Rybka 5 to surpass Houdini 2.0. Unfortunately I don't think Vas will do that; he is saving his major improvements for the Cluster version only.

Regards,
Gaмßito.
Parent - - By Bouddha (****) Date 2011-08-19 09:50
While I agree with you that it is likely that Houdini 2 will be stronger than Rybka 5, and maybe even Houdini 1.5a will be stronger than Rybka 5, I doubt that Houdini 2 will be 50 ELO points stronger than 1.5a

regards
Parent - - By Magnus Friedmann (***) Date 2011-08-19 10:13 Edited 2011-08-21 03:01
Indeed, it will not be 50 ELO points stronger, but 100!! And you know why I intend to believe it? I heard a few weeks before that Houdini 2 beta is already 60-70 points ELO than Houdini 1.5a!! When Houdini 2 will be released I think it will be 100 ELO stronger than its predecessor. Maybe we are getting closer to what Chessbase said to Garry Kasparov regarding the new thing they would bring in this autumn. Of course, Houdini 2 will be released under the new Fritz 13 interface! Robert Houdart himself wants to surpass Rybka Cluster. So Houdini 2 on 32 Cores will cause headaches to Rybka Cluster. Besides, Thinker 5.5 is beating Houdini 1.5a and Deep Rybka 4.1 running on weaker hardware than Rybka and Houdini. So, at the moment, it is very possible to be an unofficial number one in person of ........THINKER! Anyway, because it will support Chess 960, Houdini 2 will crush Critter 1.2, although the latter is already very strong. I foresee Houdini will reach 3500 ELO on quad in the near future! All the best :)
Parent - - By Carl Bicknell (*****) Date 2011-08-19 13:26

> I heard a few weeks before that Houdini 2 beta is already 60-70 points ELO than Houdini 1.5a!! When Houdini 2 will be released I think it will be 100 ELO stronger than its predecessor.


Really? Who did you hear this from? Robert is quite secretive with his testing, I wouldn't trust anyone else's word other than his. I find +100 VERY hard to believe, but if you're right...that would be incredible.
Parent - By Magnus Friedmann (***) Date 2011-08-19 15:43
I read it on a chess forum, probably on talkchess, I can't remember exactly but and I'm telling the truth, the information didn't seem unofficial so....
Parent - By InspectorGadget (*****) Date 2011-08-22 19:19

> Really? Who did you hear this from? Robert is quite secretive with his testing, I wouldn't trust anyone else's word other than his. I find +100 VERY hard to believe, but if you're right...that would be incredible.


Don't read too much into that. I don't believe Magnus at all :grin:
Parent - By jaboo2 (**) Date 2011-09-01 15:18 Edited 2011-09-01 15:27

>> So Houdini 2 on 32 Cores will cause headaches to Rybka Cluster.


But who the Hell has the money to buy a 32 core computer? Only millionaires have so much money to spend and there are relatively few millionaires in the world. Most of the people in the world are still dirt poor. Hell, I can't even afford an eight-core computer so 32 cores is of course out of the question.
Parent - - By Uly (Gold) Date 2011-08-20 01:53
I don't care much about ELO, but apparently Houdini is getting position learning, and such a feature is so powerful that if redundancy with Rybka 4 remains high, since Rybka 4 doesn't have position learning, Rybka 4 would be going obsolete for analysis.
Parent - - By M ANSARI (*****) Date 2011-08-20 06:52
Position learning is a two edged sword, it can backfire as much as can give a positive result.  I think it should never be on by default, but should be something that can be switched ON if the user wants it on.  As a user, you will know that the position learning is running and thus you can take appropriate action to prevent the negative consequences.
Parent - - By tomgdrums (****) Date 2011-08-20 11:13
What are the possible negative consequences of position learning with regards to analysis.  (not engine vs engine)

Thanks!
Parent - - By M ANSARI (*****) Date 2011-08-20 12:05
Well for one, if you have an inferior opponent, the engine might go into a position that is losing ... yet it will win the position simply because the other engine is too weak and will not play optimally the position.  This will pollute the data pool of the engine and it might "learn" that a position is good when it actually is bad.  The result would be that your engine will play weaker.
Parent - - By Uly (Gold) Date 2011-08-20 12:45
You're confusing it with book learning, which is entirely different.

The position learning we're talking about just remembers the scores of positions and depth reached last time the position was visited, and of future moves coming from the position, so that the next time you're on the position, you reach the old depth faster (you don't need to check the main PV again), with a more accurate score (because the information of future positions is backward propagated) and possibly a better move (if it turned out the played move was bad, another with a better score at earlier depth would replace it).

Game results aren't taken into account in this.

So, unless the learning is bugged (like Rybka 3 was) learning can't have any negative effects on games besides slowdown (but the slowdown should be overcame by the learning).

Now, due to the nature of chess and books, it's very improbable that the learning of old games will give better results on new games, just because playing the same out of book position is going to be very rare. This is a feature for analysis and I agree that for games it can basically be turned off without much impact.

Disclaimer - No idea how Houdini Learning is going to work, what I said applies to Shredder, and possibly future Critter, and I assume Houdini is going to do it optimally.
Parent - - By tomgdrums (****) Date 2011-08-20 13:45

> You're confusing it with book learning, which is entirely different.
>
> The position learning we're talking about just remembers the scores of positions and depth reached last time the position was visited, and of future moves coming from the position, so that the next time you're on the position, you reach the old depth faster (you don't need to check the main PV again), with a more accurate score (because the information of future positions is backward propagated) and possibly a better move (if it turned out the played move was bad, another with a better score at earlier depth would replace it).
>
> Game results aren't taken into account in this.
>
> So, unless the learning is bugged (like Rybka 3 was) learning can't have any negative effects on games besides slowdown (but the slowdown should be overcame by the learning).
>
> Now, due to the nature of chess and books, it's very improbable that the learning of old games will give better results on new games, just because playing the same out of book position is going to be very rare. This is a feature for analysis and I agree that for games it can basically be turned off without much impact.
>
> Disclaimer - No idea how Houdini Learning is going to work, what I said applies to Shredder, and possibly future Critter, and I assume Houdini is going to do it optimally.


So position learning IS good for analysis?  (which is what I am more concerned with anyway)

If I understand correctly, Shredder keeps a file of scores of positions so that even when I start up the engine a day later if it comes to a familiar position it remembers the evaluation?

Where as other programs without learning, the hash is just for analysis during that time.  It is lost when the engine is shut down?

Thanks!
Parent - By Uly (Gold) Date 2011-08-20 23:54

> So position learning IS good for analysis?  (which is what I am more concerned with anyway)


Yes, I reckon is the most effective analysis feature, and the reason Rybka 3 is still very useful to me today.

>If I understand correctly, Shredder keeps a file of scores of positions so that even when I start up the engine a day later if it comes to a familiar position it remembers the evaluation?


Yes. Problem with Shredder is that it'll keep the file very small, so that very soon, you'll have your old analysis overwritten with new analysis.

>Where as other programs without learning, the hash is just for analysis during that time.  It is lost when the engine is shut down?


Yes, and, anyway, saving the hash is very ineffective size-wise because it's saving a lot of information that isn't useful in speeding up analysis (because it contains information of positions you'll never visit), and you can't combine hashes of different games because the last one will overwrite the contents of the old game. That's why you don't see people talking about using the Save/Load Hash feature of Rybka 4, though I think Houdini will implement both methods.
Parent - By Magnus Friedmann (***) Date 2011-08-19 09:58
Thanks a lot :) for the provided information! I thought Vas cancelled Rybka 5 due to Rybka's interdiction; I'm eager for Deep Rybka 5 and I expect at least 60-70 ELO points improvement over Deep Rybka 4.1, a more agressive style, a better time management, and of course- no bugs and mosquitoes :) There has been more than a year since the release of Deep Rybka 4 and I strongly believe Vas will show us a new Rybka, a new contender for the number 1 spot. I guess he has the resources to release a much stronger version, as the Rybka Cluster is already way ahead of what he will give us by the end of the year: Deep Rybka 5!
Parent - - By Quapsel (****) Date 2011-08-19 16:13

> he is saving his major improvements for the Cluster version only.


And produces the cluster that interest that Vas had hoped for?
I had expected to read more about it, about strength, about users, about practice, about successes.
And what happened?
*  4.1 came (and that was great!)
*  ICGA-story came (and that all was not nice!)
*  and discussion in Rybka forum take place about this and that and boredom in general.

But:
What happens with the cluster today?

Quap
Up Topic Rybka Support & Discussion / Rybka Discussion / Deep Rybka 5- the project continues?
1 2 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill