err... run less programs at the same time to maximise CPU usage (NOTE: im not a computer expert at all :P)
Deep Rybka 4
Sort:
Jul 8, 2010
0
#3
Lol I actually am a bit of a computer techy but I have no idea how to use these engines... stockfish has never done me wrong :P
I don't know either -- you could try the Rybka forums, those people are chess engine techies 
http://chessprogramming.wikispaces.com/Nodes+per+second
Not sure if that helps any , but comparing between two different programs 'nodes/second' isn't that easy, and I also get similar results between Stockfish and Rybka.
If I recall, Rybka was designed to spend more of it's time in evaluating positions more accurately, whereas programs like Stockfish are more geared to blasting though many positions quickly, spending less time in the eval functions.
PeskyGnat wrote:
http://chessprogramming.wikispaces.com/Nodes+per+second
Not sure if that helps any , but comparing between two different programs 'nodes/second' isn't that easy, and I also get similar results between Stockfish and Rybka.
If I recall, Rybka was designed to spend more of it's time in evaluating positions more accurately, whereas programs like Stockfish are more geared to blasting though many positions quickly, spending less time in the eval functions.
85 vs 1500 for more in depth analysis still seems quite a large difference!
Yeah, it's orders of magnitude! so I'm wondering what's up, I can only guess they have a really good eval function. in the end, we don't really know for sure as Rybka is proprietary and can only guess as to where it is spending it's CPU cycles.
I played around a bit with the Stockfish source, very well written code, I couldn't get the same nodes/second compiling it with MSVC or MinGW as the pre-compiled one they released *shakes fist*
Also I thought I'd read Rybka4 was supposedly faster than 3... so the less nodes is odd. I also know Rybka 3 hid it's actual node count, and you had to be a freaking genius programmer to take it apart and reverse build it to discover the correct node count (ok maybe not that hard, but to someone like me it seems like magic lol).
Anyway just food for thought.
Catalyst_Kh wrote:
Don't worry, 80 kns means 1 million position for every 12 seconds, does it not enough for deep analyzes? 5 millions per only one minute already. The point of been faster for rybka means not kns number, but the ability to faster find correct lines and best deep plys, while 3000 kns Fritz and 5000 kns Junior just do the same job by diving into "every ply one by one" without too much evaluation of positions. The "high kns" approach faster and deaper in the endgames and in later middlegames with open queens, but in complex and early middlegames and in openings 5000 kns just useless speed, because this extra kns means only 2-4 additional plys, (12 Rybka plys vs 15 Fritz plys over one minute, for example), so Rybka will find better plys much faster. And there is the last problem - Rybka and Fritz both has blind spot for tactics with deep sacrifices, even after many hours of analyses, such engines with special tunes like Hiarcs (don't remember others) are slower than Fritz but has ability to break through those blind spots in positions with long attack and mate threats in middlegames, in positions with danger of passed pawn promotion and many other like this, where both Fritz (and all speedy engines like Junior too) and Rybka fails totally. Also there is an attempt to make fast engine and to make it with tunes for breaking throught the blind spots, but those attempts are not very successful yet (Shredder and stockfish for example), so those are just other engines and nothing more.
That's very interesting. I know professional players know which engine is best suited for different kinds of positions, but you seem to know a lot about it. Have you worked a lot with engines yourself or someone you know?
Catalyst_Kh wrote:
I used a lot of testing positions of all differend kind and game stages, for example this one: http://www.chess.com/forum/view/more-puzzles/absolutely-unique-masterpiece-by-troitskij2
Even hiarcs can't break through in first seconds, it needs some time (or you can just show him first move manually, then he will instantly find the rest ), while other engines are totally blind even after 3 move, just blank calculators, so kns number doesn't mean real speed. Because in simple positions hiarcs just too slow in "overcheching everything" mode.
Also example:
In this position no one engine can spot the only winning move (and lines later) for black. In many games you played (especially in open sicilian, KID-like, KG) could be a lot of such moves hiding, but you may never find that out even after the game.
Rybka and Stockfish find the winning line pretty fast, in a couple of seconds. I also showed it to a couple of other engines and they indeed have problems assessing the position correctly. Only after a long series of forced lines can they acknowledge Black's advantage.
Jul 9, 2010
0
#11
You shouldn't worry too much about that. This has always been a normal feature of the Rybka engine.
Vasil even admitted once that this display had only 'cosmetic value', and some experts wonder is the # is for real or is a trick to make the competition think Rybka's evaluation is outrageously superior... 
chessmates wrote:
What was the winning line for black?
I must admit, after quite a few minutes the only thing Fritz 12 gives me is Rd4 but with a huge winning advantage of:
1rq4k/5p2/p2R1PpP/2n1p3/4r2P/1PN2Q2/2P5/2K4R b - - 0 1
Analysis by Fritz 12:
1. =/+ (-0.36): 1...Td4 2.Td1 e4 3.De3 Pd3+ 4.Txd3 Txd3 5.Txd3 exd3 6.Dxd3 Dg4 7.Dd5 Tf8 8.Dd6 Te8 9.Kb2 Dxh4 10.Dd7 Tf8 11.Dc6 Dxh6 12.Dxa6 Te8 13.Da7 Kg8 14.Pd5 Dd2
2. = (0.00): 1...Pxb3+ 2.cxb3 Td4 3.Td1 Dc5 4.T6xd4 exd4 5.Kc2 dxc3 6.Td3 Da3 7.Txc3 Da2+ 8.Kc1 Td8 9.De3 Da1+ 10.Kc2 Da2+ 11.Kc1
3. = (0.00): 1...Tf4 2.De3 Dc7 3.Td5 Tc8 4.Dxe5 Dxe5 5.Txe5 Pxb3+ 6.Kb2 Pd2 7.Kc1
(Dekker, Limmen 09.07.2010)
Hm, I did let it run, and after 2 hrs more, you´re right, the second given Nxb3+ is now first and winning
Analysis by Fritz 12:
1. -+ (-2.33): 1...Pxb3+ 2.cxb3 Txb3 3.Kc2 Teb4 4.h5 e4 5.Dg3 Tb2+ 6.Kd1 T4b3 7.hxg6 Tb1+ 8.Pxb1 Txg3 9.gxf7 Df8 10.Td7
2. =/+ (-0.32): 1...Td4 2.Td1 e4 3.De3 Pd3+ 4.Txd3 Txd3 5.Txd3 exd3 6.Dxd3 Dg4 7.Dd6 Tc8 8.Kb2 Dxh4 9.Dxa6 Te8 10.Dd6 Dxh6 11.b4 De3
3. = (0.00): 1...Tf4 2.De3 Dc7 3.Thd1 Pxb3+ 4.cxb3 Txb3 5.T6d3 Tc4 6.Kc2 Db7 7.Dxe5 Tb2+ 8.Kc1 Tb1+ 9.Kd2 Tb2+ 10.Kc1
(Dekker, Limmen 09.07.2010)
Jul 9, 2010
0
#14
Dekker wrote:
Hm, I did let it run, and after 2 hrs more, you´re right, the second given Nxb3+ is now first and winning
Analysis by Fritz 12:
1. -+ (-2.33): 1...Pxb3+ 2.cxb3 Txb3 3.Kc2 Teb4 4.h5 e4 5.Dg3 Tb2+ 6.Kd1 T4b3 7.hxg6 Tb1+ 8.Pxb1 Txg3 9.gxf7 Df8 10.Td7
2. =/+ (-0.32): 1...Td4 2.Td1 e4 3.De3 Pd3+ 4.Txd3 Txd3 5.Txd3 exd3 6.Dxd3 Dg4 7.Dd6 Tc8 8.Kb2 Dxh4 9.Dxa6 Te8 10.Dd6 Dxh6 11.b4 De3
3. = (0.00): 1...Tf4 2.De3 Dc7 3.Thd1 Pxb3+ 4.cxb3 Txb3 5.T6d3 Tc4 6.Kc2 Db7 7.Dxe5 Tb2+ 8.Kc1 Tb1+ 9.Kd2 Tb2+ 10.Kc1
(Dekker, Limmen 09.07.2010)
Your engine missed 5...Qxc3 with mate in 4...
Jul 9, 2010
0
#15
I saw Nxb3+ as a strong, forcing move for Black after a few seconds without a computer engine. It removes part of White's pawn cover (removal of the guard) and decreases White's king saftey.
---
Regarding Stockfish 1.7. I have enjoyed Stockfish 1.5, but 1.7 locks my computer up. I'm using it with the Fritz 12 GUI. Any ideas?
Musikmole: You could try lowering the priority of the Stockfish engine if it's trying to consume 100% of the CPU. In Task Manager, right clicking on the Stockfish process should let you "set priority", try 'below normal' that's usually what I need to do sometimes.
Jul 9, 2010
0
#17
Musikamole wrote:
I saw Nxb3+ as a strong, forcing move for Black after a few seconds without a computer engine. It removes part of White's pawn cover (removal of the guard) and decreases White's king saftey.
---
Regarding Stockfish 1.7. I have enjoyed Stockfish 1.5, but 1.7 locks my computer up. I'm using it with the Fritz 12 GUI. Any ideas?
GEt the 1.8 version of Stockfish...
And FWIW the computer doesn't "know" that there is a winning move, whereas the human does. A computer programmed to find the winning move would investigate the winning move much earlier, whereas in fact it discards it fairly early on in the analysis. If you didn't kow there was a winning move wht is the chance that you would have found it?
orangehonda wrote:
I don't know either -- you could try the Rybka forums, those people are chess engine techies
Here is a forum from the official Rybka forums discussing the type of hardware needed to maximize Rybka.
Be ready to read through lines of low-level hardware specs, minute Rybka engine configurations and expensive recommendations.
Did you know that Deep Rybka 4 can use 2048 cores? (Topalov used a unique Rybka version for a "cluster farm" and is not available to the public.)
Soup up your hardware with their recommendation then make a comparison test with engines such as Stockfish, Fritz, Hiarcs, Shredder or Junior.
Apr 1, 2011
0
#19
You cannot compare Rybka's kN/s to another engine's kN/s. It's really sort of like comparing apples to oranges. You CAN, however, compare a specific version of one engine to the SAME version of the SAME engine on another machine. Each engine has different search algorithms and look for the best move in different ways. Compare it to searching for a lost child. Let's let one group of people wearing blue shirts, say 100,000 of them, look for that lost child in every city in the USA at the same time. Now, let's let another group of people wearing red shirts, say 10,000 of them, look at the same time, but in the city where the child was lost and in the city where his grandparents live. Which group is going to find the child first? We don't know. One engine may spend time looking like mad at every possible move, even the lame ones; and another might instead look hard at only what it feels are the good moves. Which chess engine will find the best move first? We don't know. As I understand it, Rybka might only have xxx kN/s on a basic system, but it's looking at what it feels are only good moves. I'm not touting Rybka, I'm just telling you how I understand it to be. See this article:
http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=5727
From what I understand, Rybka has patented its search alogrithms. It doesn't need to look everywhere for the little boy, just in the cities w/ the best odds.
i need something which is faster than rybka 4 . by fast i dont mean calculating the moves . as i have 8 gb ram and intel iris xe graphics ; i think my stuttering problem is related to the GUI's inabilitiy to switch fast enough between variants when i try to enter new moves. (let me explain it
actually the problem occurs after the analyze is stopped ,i try to enter the moves 1 by 1 via dragging the pieces with my mouse fastly , then the program cant match my speed . in other words it lags to enter the moves and gives me some freezes, lags etc.. btw i dont have to move them super fast even if i move it moderately slowly and just make 1 move it kinda frezzes for a sec or too . in my old pc's i dont remember having this problem but in this new one which on paper seems much better than in every aspect of hardware i had there is this problem .very weird.. ?btw dont bother to say use this engine or that cuz whichever engine i use the problem is te same even if i change the hashtable size to 1 mb the problem remains!! and 1 very pecuiliar thing is when i close my own database and open a new game or something . the lag doesnt totally disapper but decreases like %60 . (btw my database file ; ie cbh. file may not be very big database but its not divideed .let me explain that so when regular chessbase people like Daniel king. etc make files they diviide it into parts and subparts and chapters etc . so when u open the cbh file u see lots of diffrent games to click maybe 50 or 100 games .even the big databases that u can use have millions of games inside and everbody can open them properly . but i dont make my own database like that .i want everything in the same file cuz its my opening repertoire so it seems like 1 game but its a huge collection of games ,with variant after variants sometimes going till the very end of the game, openings and etc i once tried to print it ,it was like 60 pages or something..). i need a solution.please help .
Just got rybka 4!
I"d like to say it's amazing blah blah blah... but first I have to get it to work :P
When I try to analyze a position with most engines, I get a node speed of about 1500N/second. For some reason... with rybka 4 (I'm using deep rybka 4 32 bit on a dual core computer) ... I'm getting a much lower 85/second ...
Any ideas?