Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka @ 10th ICT Leiden 2010
1 2 Previous Next  
- - By Vasik Rajlich (Silver) [pl] Date 2010-05-31 17:20 Edited 2010-05-31 17:23
The 10th Leiden International CSVN Tournament was held the weekend of May 28-30 and Rybka won ahead of Deep Sjeng, Hiarcs and Deep Shredder.

A big thanks to the guys on our team for their hard work and also to the organizers for an excellent job putting together a strong and interesting field.

Rybka again ran on Lukas Cimiotti's cluster, which now contains 6 Nehalem 12-core machines and 7 Nehalem 8-core machines for a total of 13 computers and 128 cores. Aside from hardware upgrades, most of the clustering improvements from this year have centered on using the cluster for analysis rather than for competitive game play. Analysis in multi-variation mode is now especially efficient. Lukas has also spent some time improving behavior related to remote engine use, such as robustness to internet disconnections. The cluster can now safely be hosted for extended periods of time without manual assistance. Victor Zakharov and his team at Convekta deserve mention for improvements to their remote engine server/client tools.

On the opening book side, Jeroen Noomen made his long-awaited comeback. Jeroen's break was over twelve months long, an eternity in the world of opening theory. During this time, not only opening theory but also Rybka herself has changed. She now has a more dynamic style and while this is popular with users, Jeroen tends to prefer classical play. With white, Jeroen took an aggressive approach - Pandix, Komodo and Hiarcs got in trouble quickly, and only Deep Sjeng escaped the opening with an ok position. With black, Jeroen was more restrained, and Rybka was fed a diet of sound systems which led to slow, long-term play. I suspect that this was partly a matter of risk management and partly due to the fact that with black, sharp opening duels have a tendency to lead to draws. The opening play was very successful - the white opening positions were excellent and with black Rybka didn't concede any draws.

Jiri Dufek also helped Jeroen get back up to speed. I am sure there will be some questions about how things will go with two ultra-talented book guys on the same team. We have of course our plan for this, which we'll keep private for now to keep other teams guessing. :)

Our on-site operator was Hans van der Zijden.

The tournament itself was a little bit unusual. We lost the game vs Deep Sjeng, which was running on a pretty big cluster of its own. This would normally be very bad news, but Deep Sjeng went on an uncharacteristic drawing spree while Rybka kept squeezing out full points, and in the end we even won the tournament with a half-point to spare.

Tournament website:
Tournament forum thread:
Jeroen's opening report: http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=17206

Best wishes also to Larry Kaufman for a fast recovery from his illness. The latest news is that he can travel home in a week.

Vas
Parent - By oudheusa (*****) [nl] Date 2010-05-31 18:56
The cluster can now safely be hosted for extended periods of time without manual assistance.

Thanks for this pre announcement of Rybka Cluster ;)
Parent - - By Bobby C (****) Date 2010-06-01 07:02 Edited 2010-06-01 07:12

>Analysis in multi-variation mode is now especially efficient.


I am glad to see you got that working for Lukas while you left the rest of us with a broken Multi-Pv in Rybka 4(Chessbase GUI). Anytime I try to forward analyze a position then return to the root and switch to multi-pv Rybka 4 freezes up. Thanks Vas for letting us common folk know where your priority's are.
Parent - - By M ANSARI (*****) [kw] Date 2010-06-01 07:29
I use MPV quite often and haven't had a problem.  I use the CB GUI.  Is it specific to a certain position or do you have problems in all MPV analysis?
Parent - - By Bobby C (****) Date 2010-06-01 08:00

> Is it specific to a certain position or do you have problems in all MPV analysis?


I am having problems in most positions where Rybka scores one side decently favorable(~0.50-1.50) but have also had it lock up on me 8 moves into a 6.h3 Najdorf. I guess you and Lukas aren't reading the support threads because I along with a dozen others have reproduced this problem.
Parent - By bnc (***) Date 2010-06-01 12:09
I am having the same problem as Bobby C with Deep Rybka 4 and MPV analysis.  Rybka 4 frequently stalls. It happens on more than just one position.
Parent - - By Lukas Cimiotti (Bronze) [de] Date 2010-06-01 07:43
Sometimes the problem is not inside the computer but in front of it.
Multi-pv works very well in Rybka 4.
Parent - - By Uly (Gold) [mx] Date 2010-06-01 07:45
Parent - - By M ANSARI (*****) [kw] Date 2010-06-01 09:41 Edited 2010-06-01 09:48
I tried to go through all the steps, but after 5 minutes I was still on depth 17.  I guess this is one of the positions that the branching simply blows up into all sorts of lines.  Even if I would follow the steps mentioned and would be able to reproduce the bug, I really think you can't call that a blanket failure of MPV.  A more accurate description would be that in some special cases MPV doesn't work like you would like it to.  I use MPV analysis always and used it specifically when watching the Cluster games in Lieden.  Had absolutely no problems.  Maybe in some positions there is a problem or maybe when some certain steps are taken a bug can be produced, but MPV certainly works.  By the way this sounds more like the hash simply gets filled and you need to flush it.  I mean you are filling the hash tables with single PV data and then all of a sudden you switch to MPV after a lot of the tables are already filled with data from the single PV ... so what does the engine do?  I guess if you want it to discard all the single PV data and start a brand new MPV data set from scratch then you should clear the hash.  Does MPV also not work if you clear hash in that situation?  Is this bug also there if you are using a lot of RAM on your system and if you use larger hash settings?
Parent - - By Bobby C (****) Date 2010-06-01 19:41

>Is this bug also there if you are using a lot of RAM on your system and if you use larger hash settings?


I always use hash tables between 512mb-1024mb. I have 4gb of 1150mhz DDR2 on a Q9650@4.2ghz, CB GUI11, and Windows 7 Premium 64bit.  I have also had Single-PV lock up on me, in fact I sent a screenshot to Vytron when I first noticed it and he though it was "sampled search" screwing up but then we reproduced it in a GUI without sampled search. I am very pissed off that Vas is nowhere to be found and leaves all his customer support to volunteers. Ansari, any middlegame position that I have tried these steps on so far has failed.

#1) Start on any given position with single pv for about 2 minutes.
#2) Play a good centaur move on the board that is not what Rybka suggests
#3) Go back to the root and switch to multi-pv.

I do this to watch Rybka compare two moves that I am considering side by side. If I am working on position in a correspondence game sometimes just starting with multi-pv Rybka won't show the two moves that are best, I have to force Rybka to look at them with single PV and then back up and switch to multi-pv.
Parent - - By M ANSARI (*****) [kw] Date 2010-06-02 06:43 Edited 2010-06-02 06:56
When you want to switch to MPV after hash is filled with single PV search, the engine gets confused as the depths of MPV and the deep search of single PV will not match.  I guess it the algo could be designed to automatically clear hash when you switch from single PV to MPV, but really in the meantime it would take about 1 second to do it manually.  I think it is harsh to call that a bug and a better description would be to just say you would prefer you didn't have to clear hash when switching from deep SPV to MPV in some conditions.  Anyway give hash clear a try before changing from deep SPV to MPV and see if it solves the problem you were sharing.
Parent - - By Bobby C (****) Date 2010-06-02 06:56
Your telling me that Rybka fills a 1gb has in less then 2 minutes? Why would I want to use multi-pv if Rybka didn't remember the moves I showed it with single-pv?
Parent - - By M ANSARI (*****) [kw] Date 2010-06-02 07:43
Well single PV and multi PV are quite different and fill up tables differently.  Obviously a single PV will go deeper much quicker and if you have a deep single PV and you then start a multi PV, there is a possibility that due to MP non deterministic behavior of MP search that the single PV will not be as good a move as the second or third move of the MPV search.  This happens quite often ... you search a position in single variation mode and Rybka chooses a move, when searching the same position with multi variations it finds a stronger move almost immediately than a deep search of SV mode.  From what I understand from your thread, you are expecting that the SV variation is the strongest simply because it searched that variation longer than in MV mode, and you want it to automatically be the first choice of the MV variation ... that is simply wrong and would not be accurate.   That is not how it works and you HAVE to clear hash when the hash table is filled up with SV mode because the hash tables will be generated totally differently.  Of course if would be nice if Rybka 4 would automatically understand this and clear the hash automatically, but most likely the generated hash from SV mode would have been useful for persistent hash usage which was removed from the final version.  Anyway if you want to do what you were explaining in your thread, just clear hash before switching to MV mode.  It would take about a second or two.
Parent - - By Uly (Gold) [mx] Date 2010-06-02 07:48
Other engines are capable of remembering hash contents of SPV moves when going to MPV, Rybka may stall for 40 hours, it seems like a clear bug.
Parent - - By Dragon Mist (****) [hr] Date 2010-06-02 23:12
This is a clear bug. A huge bug, no doubt about it.
Parent - - By Uly (Gold) [mx] Date 2010-06-03 11:08
That's what I've been saying, some members cannot even do they usual analysis, as the engine would stall.

Another proof of the bug is that after hitting, the nodes per second reported will sky-rocket, to the high 800, etc. which is clearly abnormal behavior.
Parent - - By KT (**) Date 2010-06-03 12:40
Thank you guys for finding this bug. I will purchase as soon as this bug is fixed.
Parent - - By Mark (****) [us] Date 2010-06-03 13:36
You'll probably have to wait a couple years for Rybka 5, though!
Parent - - By KT (**) Date 2010-06-03 15:08
What's another year ...or two. :-)
Parent - - By InspectorGadget (*****) [za] Date 2010-06-03 16:45

> What's another year ..


Most likely a year :)
Parent - - By Mark (****) [us] Date 2010-06-03 17:31
Are you taking bets??  :)
Parent - By InspectorGadget (*****) [za] Date 2010-06-03 17:33
No :)
Parent - - By staylor (***) [il] Date 2010-06-04 11:11
Rybka 5 might only come when Vas comes crying that his online project didn't work.
Otherwise,, we might not see a Rybka 5.It was enough hard to get out this terrible Rybka 4, like squeezing water from a stone.

However, perhaps Rybka 5 might come more easily. as soon as his online gets going well. In fact, I'm sure that Rybka 5 is here already, perhaps has been for a long time. It needs about 4 days for beta testing, and then it will be on the market.

There might just be a few bugs, to make people buy Rybka 6.
Parent - - By Mark (****) [us] Date 2010-06-04 11:39

> Rybka 5 might only come when Vas comes crying that his online project didn't work.


I don't know.  Things are going to get really interesting when the online project gets underway!  It'll be exciting just to see how it's set up, the level of interest, whether it works or not, etc.
Parent - - By KT (**) Date 2010-06-04 11:49
I personally look really forward to online Rybka. I plan to buy n instances of the cheapest UCI compatible version to use with IDeA interactive analysis. The renting time will be short (eg. one hour). For over night analysis I will still use my old R3 (or perhaps R4) if it proves better than R3 for IDeA.
Parent - - By Mark (****) [us] Date 2010-06-04 11:53
This is what I don't understand regarding the usefulness of the online Rybka.  If you only buy say an hours worth of time, how could that analysis possibly be better than a couple overnight runs of R3 or R4?
Parent - - By KT (**) Date 2010-06-04 12:08
I am glad you asked. The keyword here is 'interactive'. When I work interactively with IDeA I want as much analysis power as possible. If you ever try IDeA with many remote egines (eg. 20+) you will know what I mean.
Parent - By Mark (****) [us] Date 2010-06-04 12:32
I've used IDeA a lot, but never with remote engines.
Parent - - By staylor (***) [il] Date 2010-06-04 11:57
"going to get really interesting"?

I'M not at all looking forward to it. I will not be able to be a part of it, due to cost, and i would be feeling that my own wishes for a great Rybka 4 have been hijacked, for the online version.

If (a personal) Rybka 4 was full strength, then I would be an enthusiastic spectator, online too.

Otherwise, I'm really not i nterested. I just want my own thing, the strongest i can get, and then i will get on with my own life and interests.
Parent - By Mark (****) [us] Date 2010-06-04 12:35

> I will not be able to be a part of it, due to cost


I doubt I will either, primarily because I don't really need it.  It's not like I'm a professional chess player or anything...
Parent - - By Bouddha (****) [ch] Date 2010-06-04 12:57

> I'M not at all looking forward to it. I will not be able to be a part of it, due to cost, and i would be feeling that my own wishes for a great Rybka 4 have been hijacked, for the online version.


Dont Worry,

I am sure there will very soon be a free Stockfish stronger than Rybka 4 available.

As per will it be stronger than the could Rybka ? I have no clue.

At least, looking at the Cluster Rybka playing, it seems amazinlgy strong.
Parent - - By staylor (***) [il] Date 2010-06-04 15:29 Edited 2010-06-04 15:42
I would be very interested if the cluster (or rybka 4 or stockfish etc.) would play Spassky's famous (11?)...h5 vs Larsen, with a big plus. I'd be even more astounded if it plays the move before that (10?)....Ng4. If it does, I'd wish to know what it coniders the best replies for White (Larsen). OR if the engine has a stronger continuation that Spassky.
This was absolute genius GM play, and i'm waiting for a machine to do as well.

Larsen Spassky, Belgrade, either 1970 or 1971.
Parent - - By Dadi Jonsson (Silver) [is] Date 2010-06-04 19:43
2kr3r/ppp1qppp/2p2n2/2b2b2/2P1pP2/1P2P3/PBQPB1PP/RN2K2R b KQ f3


Not a cluster, but Rybka 4 w32 (single core), 128MB hash. Always Score Main Move, otherwise default settings.

[-0.75]  d=16  10...Ng4 11.g3 h5 12.h3 h4 13.Bxg4 Bxg4 14.hxg4 hxg3 15.Rg1 Rh1 (0:04:01) 7549kN
[-0.63]  d=15  10...Ng4 11.g3 h5 12.h3 Nf6 13.a3 Qe6 14.Bxf6 Qxf6 15.Nc3 Kb8 16.b4 Be7 17.c5 Qe6 18.Rc1 (0:01:27) 2585kN
[-0.63]  d=14  10...Ng4 11.g3 h5 12.h3 Nf6 13.a3 Qe6 14.Bxf6 Qxf6 15.Nc3 Kb8 16.b4 Be7 17.c5 Qe6 18.Rc1 (0:01:20) 2393kN
[-0.62]  d=14  10...Ng4 11.g3 h5 12.Bxg7 Rhg8 13.Qb2 Be6 14.Bxg4 Bxg4 15.Bf6 Qd7 16.Bxd8 Rxd8 (0:01:05) 1911kN
[-0.38]  d=14  10...Rhg8 11.O–O Qe6 12.Nc3 Kb8 13.Na4 Be7 14.Bd4 g5 15.Kh1 h5 16.fxg5 Rxg5 17.Rad1 Rdg8 18.g3 Bh3 19.Rf2 (0:00:51) 1405kN
[-0.50]  d=13  10...Rhg8 11.Nc3 g5 12.O–O Bd6 13.fxg5 Rxg5 14.Na4 Kb8 15.c5 Bh3 16.Rf2 Rdg8 17.g3 Be5 18.Bxe5 Qxe5 19.Rb1 (0:00:33) 898kN
[-0.51]  d=12  10...Rhg8 11.O–O Ng4 12.Bxg4 Bxg4 13.Nc3 Rd3 14.Rf2 Rgd8 15.h3 Bf5 16.Re1 Kb8 17.Na4 Bb4 18.Bd4 (0:00:22) 574kN
[-0.51]  d=11  10...Rhg8 11.O–O Ng4 12.Bxg4 Bxg4 13.Nc3 Rd3 14.Rf2 Rgd8 15.h3 Bf5 16.Re1 Kb8 17.Na4 Bb4 18.Bd4 (0:00:20) 516kN
[-0.54]  d=10  10...Qd7 11.O–O h5 12.Rd1 h4 13.h3 Kb8 14.Be5 Rhe8 15.Nc3 Qe6 (0:00:08) 220kN
[-0.57]  d=9  10...Ng4 11.g3 Nf6 12.Bxf6 Qxf6 13.Nc3 h5 14.a3 Be7 15.b4 Qe6 16.Kf2 Kb8 17.c5 (0:00:04) 116kN
[-0.62]  d=9  10...Ng4 11.O–O Kb8 12.Bxg4 Bxg4 13.f5 Be2 14.f6 gxf6 15.Bxf6 (0:00:03) 79kN
Parent - - By maio [cz] Date 2010-06-06 22:58 Edited 2010-06-06 23:01
What I find more interesting is move 11...
There Stockfish always picks h7-h5 instantly, seeing it a clear win as it reaches greater depth.
Rybka (Deep Rybka 4) is another story. Most of the times, it picks h7-h5 early on, only to abandon it at or around the depth of 10 (rarely and randomly keeping it). At depth 15 or 16 it either goes back to h7-h5 or, most of the times, for Rh8-g8. Conclusion is, that at the depth of 18, it either has -1.44 h7-h5 move or -1.03 Rh8-g8 one according to its own calculation.

While at the depth of 19 it quite compensates with -1.30 Rd8-d3 vs. -1.51 h7-h5, this "depth mismatch" is still a quite disturbing discovery, if I may say so.
Parent - - By Dadi Jonsson (Silver) [is] Date 2010-06-06 23:28
Using the same Rybka 4 version as above, it sticks to h5 from depth 12 on:

2kr3r/ppp1qppp/2p5/2b2b2/2P1pPn1/1P2P1P1/PBQPB2P/RN2K2R b KQ -


[-1.35]  d=19  2...h5 3.h4 Rh6 4.a3 Rg6 5.Bxg4 Bxg4 6.b4 Bf3 7.Rh2 Bd6 8.c5 Rxg3 (0:23:19) 49183kN
[-1.23]  d=18  2...h5 3.h4 Qd6 4.Rg1 Kb8 5.Rg2 Qg6 6.Qc3 Nf6 7.Ba3 Bxa3 8.Nxa3 Bg4 9.Bxg4 Qxg4 10.Kf1 Qh3 11.Kg1 Rh6 (0:13:28) 27465kN
[-1.23]  d=17  2...h5 3.h4 Qd6 4.Rg1 Kb8 5.Rg2 Qg6 6.Qc3 Nf6 7.Ba3 Bxa3 8.Nxa3 Bg4 9.Bxg4 Qxg4 10.Kf1 Qh3 11.Kg1 Rh6 (0:09:18) 18858kN
[-1.20]  d=16  2...h5 3.h4 Qd6 4.Rg1 Kb8 5.Bxg7 Rhg8 6.Bxg4 Bxg4 7.Be5 Qe6 8.Nc3 Bf3 (0:06:30) 13390kN
[-1.22]  d=15  2...h5 3.Bxg7 Rh7 4.Qc3 Bb4 5.Bxg4 Bxg4 6.Qe5 Qd7 7.O–O Bh3 8.Bf6 Re8 9.Qd4 Bxf1 10.Qxd7+ Kxd7 11.Kxf1 Ke6 12.Bd4 Kf5 13.a3 Bd6 14.c5 Be7 15.Nc3 (0:03:24) 6914kN
[-0.96]  d=15  2...h5 3.h3 h4 4.Bxg4 Bxg4 5.hxg4 hxg3 6.Rg1 Rh2 7.Qc3 Qh4 8.Qxg7 g2+ 9.Ke2 Rh1 (0:01:53) 3792kN
[-0.81]  d=15  2...h5 3.h3 h4 4.Bxg4 Bxg4 5.hxg4 hxg3 6.Rg1 Rh2 (0:01:40) 3209kN
[-0.69]  d=14  2...h5 3.h3 Nf6 4.a3 Qe6 5.Bxf6 Qxf6 6.Nc3 Be7 7.b4 Kb8 8.O–O–O c5 9.g4 hxg4 10.hxg4 Rxh1 11.Rxh1 (0:01:11) 2177kN
[-0.69]  d=13  2...h5 3.h3 Nf6 4.a3 Qe6 5.Bxf6 Qxf6 6.Nc3 Be7 7.b4 Kb8 8.O–O–O c5 9.g4 hxg4 10.hxg4 Rxh1 11.Rxh1 (0:00:42) 1224kN
[-0.58]  d=12  2...h5 3.h3 Nf6 4.a3 Bd6 5.b4 c5 6.Bxf6 Qxf6 7.Nc3 Kb8 8.Kf2 g5 9.Nd5 Qg7 10.b5 gxf4 (0:00:32) 936kN
[-0.57]  d=12  2...h5 3.Bxg7 Rh7 4.Qb2 h4 5.Bxg4 Bxg4 6.Bf6 Qd6 7.Bxd8 hxg3 8.Bf6 Rxh2 9.Rxh2 gxh2 (0:00:29) 832kN
[-0.46]  d=12  2...Rhg8 3.h3 Nh6 4.Nc3 g5 5.O–O–O gxf4 6.gxf4 Kb8 7.Rdg1 a6 8.Kb1 Bb4 9.Ka1 Rg6 10.Kb1 Qe6 11.Rg5 (0:00:27) 758kN
[-0.45]  d=11  2...Rhg8 3.h3 Nf6 4.g4 Nh5 5.gxh5 Qh4+ 6.Kf1 g5 7.hxg6 Bxh3+ 8.Rxh3 Qxh3+ 9.Ke1 Qg3+ (0:00:24) 695kN
[-0.67]  d=11  2...Rhg8 3.h3 Nf6 4.g4 Nh5 5.gxh5 Qh4+ 6.Kf1 g5 7.hxg6 Bxh3+ 8.Rxh3 Qxh3+ 9.Ke1 Qg3+ (0:00:23) 665kN
[-0.52]  d=11  2...Rhg8 3.h3 Nf6 4.g4 Bg6 5.Nc3 h6 6.O–O–O Kb8 7.Kb1 (0:00:16) 448kN
[-0.36]  d=11  2...h5 3.Bxg7 Rhg8 4.Qb2 Qd6 5.Bc3 Qg6 6.Na3 Be7 7.Nc2 Nxh2 8.O–O–O Nf3 9.Nd4 Nxd4 10.Bxd4 c5 11.Bxh5 Qxg3 (0:00:08) 229kN
Parent - - By maio [cz] Date 2010-06-07 08:38
I read more about it since, and it is a confirmed "luck factor" by the author. It is of dubious (read negative) value for analysis, but author claims it has no effect on R4 playing strength (which seems to be confirmed by chess engine rating lists).

Make sure to clear R4 hash and run analysis again (you won't clear hash-table data stored in memory by simply closing the engine; you must either use some memory management utility or restart Windows; or at least let another engine run analysis so that it hopefully over-writes R4's data left in the memory).

To be frank, the first evening I ran the analysis, I got the results very similar to yours (R4 abandoned h5 move only to return to it later). But since then I am unable to reproduce the same result twice. And that is what was and remains a troublesome if now confirmed feature of R4 to me.

Link: http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?pid=251429#pid251429
Parent - - By Dadi Jonsson (Silver) [is] Date 2010-06-07 08:45
The "luck factor" is nothing specific to Rybka. Try other multiprocessor engines and you will see.
Parent - By maio [cz] Date 2010-06-07 09:27 Edited 2010-06-07 09:37
OK, theoretical point taken.
But in practice, when I have to choose between Stockfish that has always a "good luck" to pick h7-h5 while R4 only randomly and rarely in that... there seems to be something fishy, even if both engines have something ichthyic to their names :-)
Attachment: Fishy.pgn - Look at the last few lines of 11...h7-h5 calculations (10k)
Parent - - By Mark (****) [us] Date 2010-06-08 01:23
I've been running this on IDeA for a couple days (d=15 and 30 seconds). 

11. -- h5 is at -2.38, as is Rd3

Also, after 10. -- Ng4, IDeA likes 11. Qc1 (-1.22) instead of 11. g3 (-2.38)
Parent - By Christian Packi (****) [de] Date 2010-06-08 17:22
Stockfish likes 11. Qc1 too at depth 29. Its still lost though.
Parent - By Christian Packi (****) [de] Date 2010-06-04 20:12
Stockfish 1.7.1

(-1.77)  d18  1...Sg4 (1/46)

1...Sg4 2.g3 h5 3.h3 h4 4.hxg4 hxg3 5.Txh8 Txh8 6.Lf1 Lxg4 7.Lxg7 Th1
Parent - By Banned for Life (Gold) Date 2010-06-03 16:54
after hitting, the nodes per second reported will sky-rocket, to the high 800, etc. which is clearly abnormal behavior.

I recall hearing that Rybka does not count hash hits in its node per second count. If this is true, than the case where Rybka is getting no hash hits while doing MPV analysis because the hash is full of deep analysis positions from a single PV would result in higher kn/s, even though much less progress is being made because the transposition table is totally dysfunctional in these cases.
Parent - By yanquis1972 (****) [us] Date 2010-06-02 15:11
how does one clear has in CB GUI w/out restarting the engine?
Parent - - By Banned for Life (Gold) Date 2010-06-03 16:47
Clearing the hash is NOT an option. This would throw away all the advantages of interactive MPV analysis. It seems likely that the hash is full of positions that have been analyzed to very deep depths that are preventing other PV lines from storing shallower depths. Dr. Mueller suggested an approach to prevent this from happening a few years ago in discussions with Vas. It involves leaving a number of slots in each hash set open for lower depths.

And this has absolutely nothing to do with MPV non-deterministic behavior.
Parent - By Vasik Rajlich (Silver) [pl] Date 2010-06-04 15:26

> Clearing the hash is NOT an option. This would throw away all the advantages of interactive MPV analysis. It seems likely that the hash is full of positions that have been analyzed to very deep depths that are preventing other PV lines from storing shallower depths. Dr. Mueller suggested an approach to prevent this from happening a few years ago in discussions with Vas. It involves leaving a number of slots in each hash set open for lower depths.


Assuming this behavior is typical (I'll need to check that first), the problem would be something more crude than that.

Vas
Parent - - By Banned for Life (Gold) Date 2010-06-01 23:35
I think you are probably right and the deep single PV has filled up the hash and is freezing out the secondary MPV lines, making them take forever to calculate. This might require a change in the algorithm that decides which entries get to stay in hash, and this might need to change when you go into MPV mode. If this is the case, using a larger hash won't prevent the problem, it will just keep it from happening until a little bit later.
Parent - - By Vasik Rajlich (Silver) [pl] Date 2010-06-04 15:28
Every new entry is always written (or at least should be).

Vas
Parent - - By Banned for Life (Gold) Date 2010-06-04 16:18
If entries at shallow depth are always written, regardless of the age/depth of what is already in the hash, then either you must be very selective about which new entries are being written, or new entries keep replacing the most recently written entry (which makes the most recent entry of questionable use unless you expect it to be needed immediately). This seems to be a MRU replacement strategy. While it is possible to create scenarios where this is optimal, they tend to be pretty unrealistic...

This is another of the many issues which should be resolved based on a study of distributions rather than with a seat of the pants guestimate:
1) The distribution of read and write requests to the hash, and
2) The distribution of times required to recalculate an entry, based on depth, etc. (i.e. the savings from finding the entry in hash)

Do you know the distribution of how long an entry will be in hash before it is read, before it is overwritten?

It should be possible to come up with good metrics that maximize the probability of finding an entry in hash times the time saved by finding it in hash. I have this suspicion that it's done by trying out a few possibilities in a vast space, and choosing the one that seems to work best.
Parent - - By Vasik Rajlich (Silver) [pl] Date 2010-06-04 17:32

> If entries at shallow depth are always written, regardless of the age/depth of what is already in the hash, then either you must be very selective about which new entries are being written, or new entries keep replacing the most recently written entry (which makes the most recent entry of questionable use unless you expect it to be needed immediately). This seems to be a MRU replacement strategy. While it is possible to create scenarios where this is optimal, they tend to be pretty unrealistic...


My hash is 4-way associative and every new entry is written. You could say that 75% of my hash is ordered by depth and 25% by age (or MRU as you put it). I am sure it's not optimal but it doesn't really matter much. Try doubling your hash size once and notice the (lack of) improvement.

> 1) The distribution of read and write requests to the hash, and


Comparable.

> 2) The distribution of times required to recalculate an entry, based on depth, etc. (i.e. the savings from finding the entry in hash)


Something like 2^depth.

> Do you know the distribution of how long an entry will be in hash before it is read, before it is overwritten?


No, I never measured this. We're talking about a couple of Elo difference at most here.

> It should be possible to come up with good metrics that maximize the probability of finding an entry in hash times the time saved by finding it in hash. I have this suspicion that it's done by trying out a few possibilities in a vast space, and choosing the one that seems to work best.


The one thing I did test was associativity, mainly because it's easy to test. 4-way was clearly better than 2-way and 8-way didn't really help much (possibly because it shrinks the MRU part of the hash).

Anyway, this stuff is a lot more interesting in theory than it is important in practice.

Vas
Parent - - By Banned for Life (Gold) Date 2010-06-05 03:29
I'm not a big theory guy, but here's what I see in practice:

TEST 1
ENGINE:               Shallow Rybka
OPTIONS:             Always Score Main Move, Display Upper Bounds
MPV:                     7
GUI:                      GUI10
POSITION:            Opening Position
ANALYSIS TIME:    1, 7, and 20 Minutes on Q6600 @ 3.55 GHz
VARIABLE:             Hash Size (4, 8, 16, 32, 64, 128, 256, 512, 1024 MB)
REPEATABLE?        Yes
DATA:                    See Below

Results Summary

HASH SIZE (MB)       DEPTH @ 1 Minutes      DEPTH @ 7 Minutes      DEPTH @ 20 Minutes
4                              15-1/7                         16-2/7
8                              15-1/7                         15-4/7
16                            15                                18
32                            15                                17-4/7
64                            15                                17-5/7                          19-3/7
128                          15                                18
256                          15-4/7                         18
512                          15-5/7                         18-2/7
1024                        15-5/7                         18-6/7
2048                        16                                18-2/7                          19-5/7

The data shows only about 1 ply of difference between largest and smallest hash sizes at 1 minute but a 1-1/4 ply variation in depth reached during the seven minute run (neglecting the 4 and 8 MB hash sizes). Similar results would probably result from other positions. The high variance around the trend line is due to the small set size, which allows important results to be lost even when the hash is not very full. Given the variance in depth as a function of time due to hash size alone, it is reasonable to assume that a better hash policy could net at least a half ply of depth, since it's clear that results are being lost even during short time periods with a large hash.

Changing gears, I guess if you're 4-way set associative, you're back with 4 16-byte hash entries. This would mean that Rybka will continue to run a little faster on i7 systems with one memory channel as opposed to two or three (as Lukas confirmed for the R3 case).

I came across the case where the kn/s go up and no more progress is made a few times as well...

DATA RUNS

New game - Rybka 4 x64, Hash Size = 4 MB, Results after 1 minute
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.11): 1.Nf3 Nf6 2.d4 d5 3.Bf4 Nc6 4.e3 Nh5 5.Bb5 Rb8 6.0-0
2. =  (0.14): 1.e4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.d4 exd4 5.Nxd4 Bc5 6.Nxc6 bxc6 7.e5 Qe7
3. =  (0.13): 1.Nc3 d5 2.d4 Nc6 3.Nf3 Bf5 4.Bf4 Nf6
4. =  (0.10): 1.d4 Nf6 2.Nf3 d5 3.Nc3 e6 4.e3 Bd6 5.Bd3 0-0
5. =  (0.06): 1.e3 Nf6 2.Nf3 d5 3.Nc3 g6 4.d4 Bg7 5.Bd2 Nc6
6. =  (0.00): 1.c4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.d4 exd4 5.Nxd4 Bc5 6.Nxc6 bxc6 7.e3 0-0 8.Bd3 d5
7. =  (0.00): 1.h3 e5 2.e4 Nf6 3.Nf3 Nc6 4.Bd3 Bc5 5.0-0

In d16

New game - Rybka 4 x64, Hash Size = 8 MB, Results after 1 minute
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.15): 1.e4 e5 2.Nc3 Nf6 3.Nf3 Nc6 4.Bc4 Bc5 5.0-0 0-0 6.d3 h6
2. =  (0.15): 1.Nc3 d5 2.d4 Nc6 3.Bf4 Bf5 4.Nf3 Nf6 5.e3 e6 6.Bd3
3. =  (0.14): 1.d4 Nf6 2.Nf3 e6 3.Nc3 d5 4.e3 Be7 5.Bd3 Nc6 6.0-0 0-0
4. =  (0.14): 1.Nf3 Nf6 2.Nc3 d5 3.d4 e6 4.e3 Be7 5.Bd3 Nc6 6.0-0 0-0
5. =  (0.12): 1.e3 g6 2.Nc3 Nf6 3.Nf3 Nc6 4.d4 d5 5.Bd3 Bg7 6.a3 0-0
6. =  (-0.04): 1.g3 e5 2.Nf3 e4 3.Nd4 c5 4.Nb3 c4 5.Nd4 Qb6 6.Nf5
7. =  (-0.05): 1.c4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.e3 Bb4 5.Qc2 0-0 6.Nd5 h6

In d16

New game - Rybka 4 x64, Hash Size = 16 MB, Results after 1 minute
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.18): 1.Nf3 Nf6 2.d4 d5 3.Bf4 Nc6 4.Nc3 Be6 5.h4 Nh5 6.Be3 Nf6 7.Qd3 a6
2. =  (0.18): 1.d4 Nf6 2.Nf3 d5 3.Bf4 Nc6 4.Nc3 Be6 5.h4 Nh5 6.Be3 Nf6 7.Qd3 a6
3. =  (0.16): 1.Nc3 Nf6 2.d4 d5 3.Bf4 c5 4.e3 cxd4 5.exd4 Qb6 6.Nb5 Na6 7.Nf3 Ne4 8.Bd3 Bd7
4. =  (0.15): 1.e4 e5 2.Nf3 Nc6 3.d4 exd4 4.Nxd4 Nf6 5.Nc3 Bb4 6.Nxc6 bxc6 7.Bd3 0-0 8.0-0 d5 9.exd5 cxd5 10.Re1 Re8 11.Rxe8+
5. =  (0.06): 1.e3 Nf6 2.Nf3 Nc6 3.d4 d5 4.Nc3 g6 5.Bd3 Bg7 6.0-0 0-0
6. =  (0.00): 1.c4 e5 2.Nc3 Nf6 3.Nf3 Nc6 4.e4 Bb4 5.Bd3 a5 6.0-0 0-0
7. =  (-0.03): 1.g3 e5 2.Nf3 e4 3.Nd4 c5 4.Nb5 a6 5.N5c3 Nf6 6.Bg2 d5 7.d3

After d15

New game - Rybka 4 x64, Hash Size = 32 MB, Results after 1 minute
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.12): 1.Nc3 Nf6 2.d4 d5 3.Nf3 e6 4.g3 Nc6 5.Bg2 Bd6 6.0-0 0-0
2. =  (0.12): 1.d4 Nf6 2.Nc3 d5 3.Nf3 e6 4.g3 Nc6 5.Bg2 Bd6 6.0-0 0-0
3. =  (0.12): 1.Nf3 Nf6 2.Nc3 d5 3.d4 e6 4.g3 Nc6 5.Bg2 Bd6 6.0-0 0-0
4. =  (0.09): 1.e3 Nf6 2.Nf3 Nc6 3.d4 d5 4.Nc3 g6 5.Bd3 Bg7 6.0-0 0-0 7.Bd2 Bd7 8.a3 a6
5. =  (0.07): 1.e4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.Bc4 Bc5 5.0-0 0-0 6.a3 a5 7.d3 d6
6. =  (-0.01): 1.c4 e5 2.Nc3 Nf6 3.Nf3 Nc6 4.d4 exd4 5.Nxd4 Bb4 6.Nxc6 bxc6 7.g3 Qe7 8.Bg2 Ba6 9.Qd3 d5
7. =  (-0.03): 1.g3 e5 2.Nf3 e4 3.Nh4 d5 4.d3 Nf6 5.dxe4 Nxe4 6.Bg2 Bc5 7.0-0 0-0 8.Nc3 Nxc3 9.bxc3 c6 10.Nf3

After d15

New game - Rybka 4 x64, Hash Size = 64 MB, Results after 1 minute
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.10): 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 4.Bc4 Bc5 5.0-0 0-0 6.d3 h6 7.a3 d6 8.Bd2
2. =  (0.10): 1.Nf3 Nf6 2.Nc3 d5 3.d4 e6 4.g3 Nc6 5.Bg2 Bd6 6.0-0 0-0 7.Be3 h6
3. =  (0.10): 1.Nc3 d5 2.d4 Nf6 3.Nf3 e6 4.g3 Nc6 5.Bg2 Bd6 6.0-0 0-0 7.Be3 h6
4. =  (0.10): 1.d4 Nf6 2.Nc3 d5 3.Nf3 e6 4.g3 Nc6 5.Bg2 Bd6 6.0-0 0-0 7.Be3 h6
5. =  (0.05): 1.e3 Nf6 2.Nf3 d5 3.d4 e6 4.Be2 Nc6 5.0-0 Bd7 6.c4 dxc4 7.Bxc4 Bd6 8.Bd3
6. =  (-0.01): 1.a3 e5 2.e3 Nf6 3.d4 exd4 4.exd4 d5 5.Nf3 Nc6 6.Nc3 Bd6 7.Bd3 0-0 8.0-0 Be6 9.h3
7. =  (-0.04): 1.c4 e5 2.Nc3 Nf6 3.Nf3 Nc6 4.e4 Bb4 5.Bd3 Bc5 6.0-0 0-0 7.h3 a5 8.Na4

After d15

New game - Rybka 4 x64, Hash Size = 128 MB, Results after 1 minute
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.20): 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 4.d4 exd4 5.Nxd4 Bb4 6.Nxc6 bxc6 7.Bd3 0-0 8.0-0 Qe7 9.Re1 Bxc3 10.bxc3 Qe5 11.Rb1
2. =  (0.14): 1.d4 d5 2.Nc3 Nf6 3.Bf4 g6 4.Nf3 Bg7 5.g3 0-0 6.Bg2 Nh5 7.Bg5 Nc6 8.0-0 Bf5 9.Qd2
3. =  (0.14): 1.Nc3 d5 2.d4 Nf6 3.Bf4 g6 4.Nf3 Bg7 5.g3 0-0 6.Bg2 Nh5 7.Bg5 Nc6 8.0-0 Bf5
4. =  (0.11): 1.Nf3 Nf6 2.d4 d5 3.Nc3 e6 4.g3 Bd6 5.Bg2 0-0 6.Nb5 Nc6 7.Nxd6
5. =  (0.04): 1.e3 e5 2.d4 e4 3.c4 Nf6 4.Nc3 Nc6 5.Nge2 Be7 6.Nf4 0-0 7.Be2
6. =  (0.01): 1.g3 e5 2.d4 exd4 3.Nf3 Nf6 4.Nxd4 d5 5.Bg2 c5 6.Nb3 a5 7.0-0 Nc6 8.a4 Bf5
7. =  (0.00): 1.h3 e5 2.e4 Nc6 3.Nc3 Nf6 4.Nf3 d5 5.exd5 Nxd5 6.Bc4 Nxc3 7.bxc3 Bc5 8.0-0 0-0 9.Qe2 Qe7 10.Re1

After d15

New game - Rybka 4 x64, Hash Size = 256 MB, Results after 1 minute
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.19): 1.Nc3 Nf6 2.d4 d5 3.Bf4 e6 4.e3 Bb4 5.Bd3 0-0 6.Nf3 Nc6 7.0-0 Bxc3 8.bxc3 h6 9.c4 Ne4 10.Ne5
2. =  (0.19): 1.d4 Nf6 2.Nc3 d5 3.Bf4 e6 4.e3 Bb4 5.Bd3 0-0 6.Nf3 Nc6 7.0-0 Bxc3 8.bxc3 h6 9.c4
3. =  (0.16): 1.Nf3 Nf6 2.d4 e6 3.c4 Be7 4.Nc3 0-0 5.e3 c5 6.Bd3 Nc6 7.0-0 d5 8.h3
4. =  (0.15): 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.Nc3 Nf6 5.0-0 0-0 6.a3 a5 7.h3 h6 8.d3 d6
5. =  (0.08): 1.e3 Nf6 2.Nf3 Nc6 3.d4 e6 4.Bd3 d5 5.0-0 Bd6 6.Nc3 0-0
6. =  (0.00): 1.c4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.e4 Bb4 5.d3 h6 6.Be2 0-0 7.0-0 d6 8.Be3 Be6 9.h3 a5 10.Qb3
7. =  (0.00): 1.h3 e5 2.e4 Nc6 3.Nc3 Nf6 4.Nf3 d5 5.exd5 Nxd5 6.Bc4 Nxc3 7.bxc3 Bc5 8.0-0 0-0 9.Qe2 Qd6 10.Qe4

In d16

New game - Rybka 4 x64, Hash Size = 512 MB, Results after 1 minute
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.14): 1.e4 e5 2.Nf3 Nc6 3.Bc4 Bc5 4.0-0 Nf6 5.Nc3 0-0 6.a3 a5 7.d3 d6 8.h3 h6 9.Bd2 Bd7
2. =  (0.12): 1.Nf3 Nf6 2.d4 d5 3.Nc3 Nc6 4.Bf4 Bf5 5.e3 e6 6.Bb5 Bd6 7.Bxd6 cxd6 8.Nh4 0-0 9.Nxf5 exf5
3. =  (0.12): 1.d4 Nf6 2.Nf3 d5 3.Nc3 Nc6 4.Bf4 Bf5 5.e3 e6 6.Bb5 Bd6 7.Bxd6 cxd6 8.Nh4 0-0 9.Nxf5 exf5
4. =  (0.12): 1.Nc3 d5 2.d4 Nf6 3.Nf3 Nc6 4.Bf4 Bf5 5.e3 e6 6.Bb5 Bd6 7.Bxd6 cxd6 8.Nh4 0-0 9.Nxf5 exf5
5. =  (0.08): 1.e3 Nf6 2.d4 d5 3.Nf3 Nc6 4.Bd3 g6 5.Nc3 Bg7 6.0-0 0-0 7.Bd2 a6 8.a3 Qd6 9.h3 e5
6. =  (0.02): 1.a3 e5 2.e3 Nf6 3.d4 exd4 4.exd4 d5 5.Nf3 Bd6 6.Qe2+ Be7 7.Nc3 0-0 8.Be3 Ne4 9.Nxe4 dxe4 10.Ne5 Nc6 11.0-0-0 Be6
7. =  (0.00): 1.c4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.d4 exd4 5.Nxd4 Bb4 6.Nxc6 bxc6 7.g3 0-0 8.Bg2 Bxc3+ 9.bxc3 Re8 10.0-0 Qe7 11.Bf3 Ne4 12.Qd3 Rb8

In d16

New game - Rybka 4 x64, Hash Size = 1 GB, Results after 1 minute
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.15): 1.Nf3 Nf6 2.c4 c5 3.Nc3 Nc6 4.e3 e5 5.Bd3 d6 6.0-0 Be7 7.Qb3 0-0 8.a3 h6 9.h3 Qb6
2. =  (0.14): 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 4.Bc4 Bc5 5.0-0 0-0 6.d3 d6 7.Na4 Be6 8.Nxc5 dxc5
3. =  (0.13): 1.d4 d5 2.Nf3 Nf6 3.Bf4 Nc6 4.Nc3 Bf5 5.e3 e6 6.Bb5 Bd6 7.Bxd6 cxd6 8.Nh4 0-0 9.Nxf5 exf5 10.0-0 Qd7 11.h3 h6 12.Qd3
4. =  (0.09): 1.Nc3 d5 2.d4 Nf6 3.Nf3 e6 4.e3 c5 5.Bd2 Be7 6.Bd3 Nc6 7.0-0 c4 8.Be2 0-0 9.Ne5 Bd7 10.Nxc6
5. =  (0.09): 1.e3 Nf6 2.Nf3 Nc6 3.d4 d5 4.Nc3 g6 5.Bd3 Bg7 6.0-0 0-0 7.h3 a6 8.Bd2 Qd6 9.Qe2 e5 10.a3
6. =  (0.00): 1.h3 e5 2.e4 Nc6 3.Nc3 Nf6 4.Nf3 d5 5.exd5 Nxd5 6.Bb5 Nxc3 7.bxc3 Bd6 8.Bxc6+ bxc6 9.0-0 0-0 10.Qe2 f6 11.Qc4+ Kh8 12.Qxc6 Be6
7. =  (0.00): 1.c4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.d4 exd4 5.Nxd4 Bb4 6.Nxc6 bxc6 7.g3 0-0 8.Bg2 Bxc3+ 9.bxc3 Qe7 10.0-0 Re8 11.Re1 Ne4

In d16

New game - Rybka 4 x64, Hash Size = 2 GB, Results after 1 minute
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.15): 1.Nc3 Nf6 2.d4 d5 3.Nf3 Bf5 4.Bf4 e6 5.e3 Bb4 6.Bd3 0-0 7.0-0 Nc6 8.Nb5 Bxd3 9.Qxd3 Rc8 10.Ne5
2. =  (0.15): 1.d4 Nf6 2.Nc3 d5 3.Nf3 Bf5 4.Bf4 e6 5.e3 Bb4 6.Bd3 0-0 7.0-0 Nc6 8.Nb5 Bxd3 9.Qxd3 Rc8 10.Ne5
3. =  (0.15): 1.Nf3 Nf6 2.Nc3 d5 3.d4 Bf5 4.Bf4 e6 5.e3 Bb4 6.Bd3 0-0 7.0-0 Nc6 8.Nb5 Bxd3 9.Qxd3 Rc8 10.Ne5
4. =  (0.15): 1.e4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.Bc4 Bc5 5.0-0 0-0 6.d3 h6 7.a3 a5 8.Be3 Bxe3 9.fxe3 d6 10.Nd5
5. =  (0.04): 1.e3 Nf6 2.Nf3 d5 3.Nc3 g6 4.d4 Bg7 5.Bd3 0-0 6.0-0 Nc6 7.h3 a6 8.Qe2 Be6 9.Bd2
6. =  (0.02): 1.c4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.e4 Bd6 5.Be2 0-0 6.0-0 h6 7.h3 Nd4 8.d3 Nxe2+ 9.Qxe2 b6 10.Nb5 Qe7 11.Nxd6 cxd6 12.Bd2 Bb7
7. =  (0.01): 1.a3 e5 2.e3 Nf6 3.d4 exd4 4.exd4 d5 5.Nf3 Bd6 6.Bd3 Qe7+ 7.Qe2 Be6 8.Nc3 Nc6 9.0-0 0-0 10.h3 Rae8 11.Re1

After d16

New game - Rybka 4 x64, Hash Size = 4 MB, Results after 7 minutes
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.16): 1.e4 e5 2.Nf3 Nc6 3.Bb5 Bd6 4.0-0 Nge7 5.Na3 Bxa3 6.bxa3 0-0 7.d4
2. =  (0.16): 1.d4 d5 2.Nc3 Nf6 3.Bf4 Bf5 4.e3 e6 5.Bd3 Nc6 6.Bxf5 exf5 7.Nge2 Bb4
3. =  (0.14): 1.Nc3 e5 2.Nf3 Nc6 3.d4 exd4 4.Nxd4 Nf6 5.e4 Bc5 6.Be3 Bxd4 7.Bxd4
4. =  (0.11): 1.Nf3 Nf6 2.d4 d5 3.Bf4 Nc6 4.e3 Nh5 5.Bb5 Rb8 6.0-0
5. =  (0.07): 1.e3 Nf6 2.Nf3 d5 3.d4 g6 4.Bd3 Bg7 5.0-0 Nc6 6.Nc3 0-0
6. =  (0.00): 1.h3 e5 2.e4 Nf6 3.Nc3 Nc6 4.Nf3 d5 5.exd5 Nxd5 6.Bc4 Nxc3 7.bxc3 e4
7. =  (-0.01): 1.c4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.d4 exd4 5.Nxd4 Bc5 6.Nxc6 bxc6 7.e3 0-0 8.Bd3 Rb8 9.0-0

In d17

New game - Rybka 4 x64, Hash Size = 8 MB, Results after 7 minutes
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.16): 1.Nc3 Nf6 2.d4 d5 3.Nf3 e6 4.e3 Be7 5.Bd3 Nc6 6.0-0 0-0 7.a3
2. =  (0.16): 1.d4 Nf6 2.Nf3 e6 3.Nc3 d5 4.e3 Be7 5.Bd3 Nc6 6.0-0 0-0 7.a3
3. =  (0.16): 1.Nf3 Nf6 2.Nc3 d5 3.d4 e6 4.e3 Be7 5.Bd3 Nc6 6.0-0 0-0 7.a3
4. =  (0.15): 1.e4 e5 2.Nc3 Nf6 3.Nf3 Nc6 4.Bc4 Bc5 5.0-0 0-0 6.d3 h6
5. =  (0.12): 1.e3 g6 2.Nc3 Nf6 3.Nf3 Nc6 4.d4 d5 5.Bd3 Bg7 6.a3 0-0
6. =  (-0.04): 1.g3 e5 2.Nf3 e4 3.Nd4 c5 4.Nb3 c4 5.Nd4 Qb6 6.Nf5
7. =  (-0.05): 1.c4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.e3 Bb4 5.Qc2 0-0 6.Nd5 h6

In d16

New game - Rybka 4 x64, Hash Size = 16 MB, Results after 7 minutes
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.20): 1.e4 e6 2.Nc3 d5 3.Nf3 d4 4.Ne2 c5 5.c3 Nc6 6.cxd4 Nf6 7.e5 Nd5 8.Qa4 Qc7
2. =  (0.11): 1.d4 Nf6 2.Nf3 e6 3.e3 c5 4.c4 d5 5.Bd3 dxc4 6.Bxc4 Nc6 7.0-0 Be7
3. =  (0.11): 1.Nf3 Nf6 2.d4 e6 3.e3 c5 4.c4 d5 5.Bd3 dxc4 6.Bxc4 Nc6 7.0-0 Be7
4. =  (0.09): 1.e3 Nf6 2.Nf3 Nc6 3.d4 d5 4.Nc3 g6 5.Bd3 Bg7 6.0-0 0-0 7.h3 Nb4 8.Ne5
5. =  (0.08): 1.Nc3 d5 2.d4 Nf6 3.Bf4 Bf5 4.Nf3 e6 5.Nh4 Bd6 6.Nxf5 exf5 7.Bg5
6. =  (0.00): 1.g3 e5 2.Nf3 e4 3.Nh4 Nf6 4.d3 exd3 5.Qxd3 Bc5 6.Bg2 0-0 7.0-0 Nc6 8.Nc3
7. =  (0.00): 1.c4 e5 2.Nc3 Nf6 3.Nf3 Nc6 4.e4 Bb4 5.d3 0-0 6.Be2 d6 7.0-0 h6 8.Be3

After d18

New game - Rybka 4 x64, Hash Size = 32 MB, Results after 7 minutes
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.20): 1.e4 e5 2.Nf3 Nc6 3.Bb5 Bd6 4.0-0 Nge7 5.Bc4 0-0 6.Nc3 Bc5 7.d3 d6 8.Be3
2. =  (0.14): 1.d4 Nf6 2.Nf3 d5 3.g3 Nc6 4.Bg2 g6 5.Nc3 Bg7 6.0-0 Bf5 7.Bf4 Nh5
3. =  (0.14): 1.Nf3 Nf6 2.d4 d5 3.g3 Nc6 4.Bg2 g6 5.Nc3 Bg7 6.0-0 Bf5 7.Bf4 Nh5
4. =  (0.05): 1.Nc3 d5 2.d4 Nf6 3.Bf4 Bf5 4.e3 e6 5.Bd3 Nc6 6.Nf3 Bb4 7.0-0 Bxc3
5. =  (0.08): 1.e3 Nf6 2.Nf3 Nc6 3.d4 d5 4.Nc3 g6 5.Bd3 Bg7 6.0-0 Nb4 7.Ne5
6. =  (-0.02): 1.h3 e5 2.e4 Nf6 3.Nc3 d5 4.exd5 Nxd5 5.Qh5 Nc6 6.Bb5 Be7 7.Nf3 0-0 8.Bxc6 bxc6
7. =  (-0.03): 1.c4 e5 2.Nc3 Nf6 3.Nf3 Nc6 4.e4 Bb4 5.d3 0-0 6.Be2 d6 7.h3

In d18

New game - Rybka 4 x64, Hash Size = 64 MB, Results after 7 minutes
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.20): 1.e4 e6 2.Nc3 d5 3.d4 Bb4 4.exd5 exd5 5.Bd3 Nc6 6.Nge2 Nf6 7.0-0 0-0 8.Bg5 Be6
2. =  (0.17): 1.Nf3 Nf6 2.c4 c5 3.e3 Nc6 4.Nc3 e6 5.d4 d5 6.Bd3 dxc4 7.Bxc4 cxd4 8.exd4 Bd6 9.0-0
3. =  (0.17): 1.d4 Nf6 2.c4 e6 3.Nf3 d5 4.e3 c5 5.Bd3 Nc6 6.Nc3 dxc4 7.Bxc4 cxd4 8.exd4 Bd6 9.0-0
4. =  (0.17): 1.e3 Nf6 2.Nf3 d5 3.c4 e6 4.d4 c5 5.Bd3 Nc6 6.Nc3 dxc4 7.Bxc4 cxd4 8.exd4 Bd6 9.0-0
5. =  (0.11): 1.Nc3 e5 2.Nf3 Nc6 3.d4 exd4 4.Nxd4 Nf6 5.Nxc6 dxc6 6.Qxd8+ Kxd8 7.e4 Bd6 8.Be3 Ke7 9.0-0-0 Be6 10.f3 Be5 11.Be2
6. =  (0.00): 1.a3 Nf6 2.d4 d5 3.Nf3 Nc6 4.Nc3 Bf5 5.Nh4 Be6 6.Nf3 a6 7.Bf4 Ne4 8.Qd3
7. =  (-0.05): 1.h3 e5 2.e4 Nf6 3.Nc3 Bb4 4.Bd3 0-0 5.Qe2 Nc6 6.Nf3 Nd4 7.Nxd4 exd4 8.Nb5

In d18

New game - Rybka 4 x64, Hash Size = 128 MB, Results after 7 minutes
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.23): 1.e4 e5 2.Nf3 Nf6 3.d4 exd4 4.e5 Ne4 5.Qxd4 d5 6.exd6 Nxd6 7.Qe5+ Qe7 8.Qxe7+ Bxe7 9.Nc3 Nc6 10.Nd5 Bd8 11.Be3 Be6 12.0-0-0 Ne7
2. =  (0.15): 1.e3 Nf6 2.Nf3 d5 3.d4 e6 4.c4 Be7 5.Bd3 0-0 6.0-0 c5 7.Nc3 dxc4 8.Bxc4
3. =  (0.12): 1.d4 d5 2.Bf4 Bf5 3.e3 e6 4.Bd3 Nf6 5.Bxf5 exf5 6.Qd3 Nc6 7.Nc3 Ne4 8.Qb5 Nxc3 9.bxc3 Rb8 10.Nf3 Be7
4. =  (0.12): 1.Nc3 d5 2.d4 Nf6 3.Bf4 Bf5 4.e3 e6 5.Bd3 Nc6 6.Bxf5 exf5 7.Qd3 Ne4 8.Qb5 Nxc3 9.bxc3 Rb8 10.Nf3 Be7
5. =  (0.10): 1.Nf3 Nf6 2.d4 d5 3.Bf4 Bf5 4.e3 e6 5.Bd3 Bd6 6.0-0 0-0 7.Bxf5 exf5 8.Qd3 Ne4 9.Qb3
6. =  (0.01): 1.c4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.e4 Bb4 5.d3 0-0 6.Be2 h6 7.0-0 d6 8.h3 Bc5
7. =  (-0.02): 1.a3 e5 2.e3 Nc6 3.d4 d5 4.dxe5 Nxe5 5.f4 Nc6 6.Nf3 Nf6 7.Bd3 Bd6 8.0-0

After d18

New game - Rybka 4 x64, Hash Size = 256 MB, Results after 7 minutes
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.17): 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 4.d4 exd4 5.Nxd4 Bb4 6.Nxc6 bxc6 7.Bd3 0-0 8.0-0 d5 9.exd5 cxd5 10.Bf4 c6 11.Qf3 Bg4 12.Qg3
2. =  (0.15): 1.Nf3 Nf6 2.d4 d5 3.Bf4 Nc6 4.Nc3 Nh5 5.Bd2 g6 6.e3 Bg7 7.Bd3 0-0 8.0-0 e5
3. =  (0.15): 1.e3 Nf6 2.Nf3 Nc6 3.d4 e6 4.c4 Be7 5.Bd3 d5 6.Nc3 0-0 7.0-0 Nb4 8.Be2 dxc4 9.Bxc4 Nbd5 10.Bd3
4. =  (0.12): 1.d4 Nf6 2.Nc3 d5 3.Bf4 Bf5 4.e3 e6 5.Bd3 Nc6 6.Bxf5 exf5 7.Qd3 Bb4 8.Qxf5 Bxc3+ 9.bxc3 Ne4 10.Ne2 g6 11.Qh3
5. =  (0.12): 1.Nc3 d5 2.d4 Nf6 3.Bf4 Bf5 4.e3 e6 5.Bd3 Nc6 6.Bxf5 exf5 7.Qd3 Bb4 8.Qxf5 Bxc3+ 9.bxc3 Ne4 10.Ne2 g6 11.Qh3
6. =  (0.02): 1.a3 e5 2.e3 d5 3.d4 exd4 4.exd4 Bd6 5.Nf3 Nf6 6.Bd3 0-0 7.0-0 Nc6 8.Nc3 h6 9.Be3 Re8 10.h3 a6
7. =  (0.00): 1.c4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.e4 Bb4 5.d3 h6 6.Be2 0-0 7.0-0 d6 8.Be3 Be6 9.h3 a5

After d18

New game - Rybka 4 x64, Hash Size = 512 MB, Results after 7 minutes
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.18): 1.e4 e6 2.Nf3 d5 3.exd5 exd5 4.d4 Nf6 5.Bb5+ c6 6.Bd3 Qe7+ 7.Be3 Ng4 8.Qe2 Nxe3 9.fxe3 Be6 10.Nc3 Nd7 11.0-0-0 0-0-0 12.a3
2. =  (0.16): 1.Nf3 Nf6 2.d4 d5 3.e3 e6 4.c4 Be7 5.Nc3 0-0 6.Bd3 dxc4 7.Bxc4 Nc6 8.0-0 Bd7 9.Bd2 a6 10.a3
3. =  (0.13): 1.d4 Nf6 2.Nf3 d5 3.g3 Nc6 4.Bg2 g6 5.0-0 Bf5 6.Nc3 Bg7 7.Ne5 0-0 8.Bf4 Nb4 9.Rc1
4. =  (0.12): 1.Nc3 d5 2.d4 Nf6 3.Bf4 Bf5 4.e3 Nc6 5.Bd3 e6 6.Bxf5 exf5 7.Qd3 Bb4 8.Qxf5 Bxc3+ 9.bxc3 Ne4 10.Ne2 g6 11.Qh3 0-0
5. =  (0.06): 1.e3 Nf6 2.d4 g6 3.Nf3 Bg7 4.Bd3 Nc6 5.Nc3 d5 6.a3 0-0 7.h3 a6 8.0-0 Qd6 9.Bd2 Re8
6. =  (0.00): 1.a3 e5 2.e3 Nc6 3.d4 d5 4.Nf3 e4 5.Nfd2 f5 6.c4 Nf6 7.Qb3 Na5 8.Qa4+ Nc6 9.Qb3 Na5 10.Qa4+ Nc6 11.Qb3 Na5 12.Qa4+ Nc6 13.Qb3 Na5 14.Qa4+ Nc6 15.Qb3 Na5 16.Qa4+
7. =  (-0.01): 1.c4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.d4 exd4 5.Nxd4 Bc5 6.e3 0-0 7.Be2 d6 8.Nxc6 bxc6 9.0-0 Bf5 10.Qa4 Ne4 11.Qxc6 Nxc3 12.bxc3

In d19

New game - Rybka 4 x64, Hash Size = 1 GB, Results after 7 minutes
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.18): 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.0-0 Bc5 5.Nxe5 Nxe5 6.d4 a6 7.Ba4 b5 8.Bb3 Bxd4 9.Qxd4 d6 10.Nc3 c5 11.Qe3 0-0 12.f4
2. =  (0.15): 1.Nc3 d5 2.d4 Nf6 3.Bf4 Bf5 4.e3 Nc6 5.Bd3 e6 6.Bxf5 exf5 7.Qd3 Ne4 8.Qb5 Nxc3 9.bxc3 Rb8 10.Nf3 Qd7 11.0-0 a6 12.Qe2 Be7 13.c4
3. =  (0.13): 1.Nf3 d5 2.d4 Nf6 3.Bf4 Nc6 4.Nc3 Bf5 5.e3 e6 6.Bb5 Bd6 7.0-0 0-0 8.Ne5 Bxe5 9.dxe5 Nd7 10.Bxc6 bxc6 11.f3 Rb8 12.e4 Bg6
4. =  (0.13): 1.d4 d5 2.Nf3 Nf6 3.Bf4 Nc6 4.Nc3 Bf5 5.e3 e6 6.Bb5 Bd6 7.0-0 0-0 8.Ne5 Bxe5 9.dxe5 Nd7 10.Bxc6 bxc6 11.f3 Rb8 12.e4 Bg6
5. =  (0.13): 1.e3 Nf6 2.Nf3 d5 3.c4 e6 4.d4 Be7 5.Nc3 0-0 6.Bd3 c5 7.0-0 Nc6 8.cxd5 exd5 9.dxc5 Bxc5 10.a3 Be6 11.b4
6. =  (0.02): 1.c4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.d4 exd4 5.Nxd4 Bb4 6.Nxc6 bxc6 7.e3 0-0 8.Bd3 Rb8 9.0-0 d5 10.Qa4 a5 11.cxd5 cxd5 12.Bd2 Bd7
7. =  (0.01): 1.h3 Nf6 2.Nf3 d5 3.d4 e6 4.e3 c5 5.c4 cxd4 6.exd4 Bb4+ 7.Bd2 Nc6 8.Bxb4 Nxb4 9.Be2 dxc4 10.0-0 0-0 11.Nbd2 Bd7 12.Nxc4

In d19

New game - Rybka 4 x64, Hash Size = 2 GB, Results after 7 minutes
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.16): 1.e4 e6 2.Nf3 d5 3.exd5 exd5 4.d4 Nf6 5.Bd3 c5 6.0-0 c4 7.Be2 Be7 8.Ne5 Nc6 9.Nc3 0-0 10.Bf4 Qb6 11.Nxc6 Qxc6
2. =  (0.13): 1.Nf3 Nf6 2.d4 d5 3.Bf4 g6 4.Nc3 Bg7 5.e3 Nh5 6.Bg5 Nc6 7.Bb5 f6 8.Bh4 0-0 9.0-0
3. =  (0.13): 1.d4 Nf6 2.c4 e6 3.Nf3 d5 4.e3 Be7 5.Bd3 0-0 6.0-0 c5 7.cxd5 Nxd5 8.dxc5 Nb4 9.Be2 Qxd1 10.Bxd1 Nd3 11.Nc3 Nc6 12.Nd4
4. =  (0.08): 1.e3 Nf6 2.Nf3 d5 3.d4 g6 4.Nc3 Bg7 5.Bd3 0-0 6.0-0 Nc6 7.h3 a6 8.Bd2 Qd6 9.a3 Be6 10.Re1 Rfe8
5. =  (0.07): 1.Nc3 d5 2.d4 Nf6 3.Nf3 Bf5 4.g3 Nc6 5.Nh4 Bd7 6.Bg2 e6 7.0-0 Be7 8.Nf3 0-0 9.Bf4 h6 10.Nb5 Rc8 11.Qd3 a5
6. =  (0.01): 1.a3 e5 2.e3 Nc6 3.d4 d5 4.dxe5 Nxe5 5.Nc3 Nf6 6.f4 Nc6 7.Nf3 Bd6 8.Be2 0-0 9.0-0 Be6 10.Ng5 Qc8
7. =  (0.00): 1.c4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.e4 Bb4 5.d3 0-0 6.Be2 d6 7.0-0 h6 8.Be3 Be6 9.h3 a5 10.Re1 Bd7

In d19

New game - Rybka 4 x64, Hash Size = 64 MB, Results after 20 minutes
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.14): 1.e4 e6 2.Nc3 d5 3.d4 Nf6 4.exd5 exd5 5.Bd3 Nc6 6.Nf3 Bb4 7.Qe2+ Be6 8.0-0 0-0 9.Re1 Re8
2. =  (0.11): 1.Nf3 Nf6 2.c4 e6 3.d4 d5 4.e3 c5 5.Bd3 dxc4 6.Bxc4 Nc6 7.0-0 Be7 8.dxc5 0-0 9.a3
3. =  (0.11): 1.d4 Nf6 2.c4 e6 3.Nf3 d5 4.e3 c5 5.Bd3 dxc4 6.Bxc4 Nc6 7.0-0 Be7 8.dxc5 0-0 9.a3
4. =  (0.16): 1.e3 Nf6 2.Nf3 d5 3.d4 e6 4.c4 c5 5.Bd3 dxc4 6.Bxc4 Nc6 7.0-0 cxd4 8.exd4 Bd6 9.Be3
5. =  (0.10): 1.Nc3 e5 2.Nf3 Nc6 3.d4 exd4 4.Nxd4 Nf6 5.e4 Bb4 6.Nxc6 bxc6 7.Bd3 0-0 8.0-0 d5 9.exd5 cxd5 10.Bf4 c6 11.Re1 Be6 12.a3 Bd6
6. =  (0.00): 1.a3 Nf6 2.d4 d5 3.Nc3 Bf5 4.Nf3 Nc6 5.g3 e6 6.Nh4 Bg4
7. =  (0.00): 1.c4 e5 2.Nc3 Nf6 3.Nf3 Nc6 4.e4 Bb4 5.d3 0-0 6.Be2 d6 7.0-0 h6 8.Be3 a5 9.h3

In d20

New game - Rybka 4 x64, Hash Size = 2 GB, Results after 20 minutes
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1


Analysis by Rybka 4 x64:

1. =  (0.20): 1.d4 Nf6 2.c4 e6 3.Nf3 d5 4.Nc3 Be7 5.e3 0-0 6.Bd3 c5 7.0-0 Nc6 8.dxc5
2. =  (0.17): 1.e4 e6 2.d4 d5 3.Nc3 Bb4 4.exd5 exd5 5.Bd3 Nf6 6.Nge2 0-0 7.0-0 Nc6 8.Bg5 Be6 9.a3 Ba5 10.Re1 h6 11.Be3 Re8
3. =  (0.16): 1.Nf3 Nf6 2.d4 d5 3.c4 dxc4 4.e3 Be6 5.Nbd2 Nc6 6.Bxc4 Bxc4 7.Nxc4 e6 8.0-0 Qd5 9.Qb3 Rd8 10.a4 Be7 11.Bd2 0-0 12.Rac1
4. =  (0.09): 1.e3 Nf6 2.d4 g6 3.Be2 Nc6 4.c4 d5 5.cxd5 Nxd5 6.Nf3 Bh6 7.0-0 0-0 8.Nc3 Nxc3 9.bxc3
5. =  (0.04): 1.Nc3 d5 2.d4 Nf6 3.Bf4 Bf5 4.e3 e6 5.Bd3 Nc6 6.Bxf5 exf5 7.Nge2 Be7 8.Be5 0-0 9.Bxf6 Bxf6 10.Nf4 Ne7 11.0-0 g6 12.g3
6. =  (0.01): 1.a3 e5 2.e3 Nc6 3.d4 d5 4.dxe5 Nxe5 5.Nc3 Nf6 6.f4 Nc6 7.Nf3 Bd6 8.Be2 0-0 9.0-0 Ne7 10.Bd2 c6 11.Nd4
7. =  (0.00): 1.c4 e5 2.Nc3 Nc6 3.Nf3 Nf6 4.e4 Bb4 5.d3 0-0 6.Be2 d6 7.0-0 h6 8.Be3 Be6 9.h3 a5 10.Re1

In d20
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka @ 10th ICT Leiden 2010
1 2 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill