Rybka's Nemesis is in the oven

[Eelco de Groot]

The price for the MP version is more than double that of the 2.2 MP only engine. But for a 100 elo increase, that is really a huge jump in strength! It's not a jump, it's not a pole vault, it's a take off! 60 Euros, it isn't cheap, but at least Alex does not erode the market of other commercial engines this way ..

I found interesting these characteristics that Alex mentions;

* More reliable (less speculative) search. No more wild score and best move swings. Naum will provide high quality position analysis you will trust.
* Improved evaluation, especially in the area of material imbalances.

I am curious how much these characteristics are actually key for getting greater strength. The first point is like what Stefan has been aiming for also with Shredder 11 and both points seem in line with what Rybka is doing. On the other hand Fritz 11 is also reported to be stronger, but I don't know if that will have effected its tactical strength. But Fritz's tactics is not necessarily all that dependent on speculative (King-attacks) play, more on its deep calculations I think. It will be very interesting to see what the playing style of the new engines will be like, whether their style has changed. Of course the stronger engine can also afford some more risk taking, that's about the infamous search gap © and all, so it is not a simple matter of taking less risks to become stronger, I do not want to imply that.

In all, many authors and their engines seem to be making big progress lately!

Eelco

[Tony Thomas]

I wonder if he figured out the secret of Rybka, a material imbalance table is the strength of Rybka according to Mr.Chrilly.

[Erik Roggenburg]

If it truly does add 100 ELO, it would move it into rareified air in my little list:

Code: Select all

    Program                          Elo    +   -   Games   Score   Av.Op.  Draws

  1 Rybka 2.3.1 mp 32-bit          : 2821   14  14  1900    73.1 %   2648   29.2 %
  2 Rybka 2.2 mp 32-bit            : 2814   15  15  1580    72.7 %   2644   30.4 %
  3 Rybka 2.3 LK mp 32-bit         : 2808   18  18  1140    73.1 %   2635   28.8 %
  4 Rybka 2.3 mp 32-bit            : 2808   18  18  1140    73.0 %   2635   30.4 %
  5 Deep Shredder 11 UCI nobk      : 2738   20  19   900    64.2 %   2637   29.6 %
  6 Zappa Mexico 2T                : 2731   19  19   900    63.3 %   2637   34.6 %
  7 HIARCS 11.1 MP UCI             : 2726   12  12  1940    59.9 %   2656   36.6 %
  8 Loop 13.5.32 2CPU              : 2724   13  13  1740    61.1 %   2645   36.9 %
  9 Deep Fritz 10                  : 2722   13  13  1980    59.2 %   2657   31.2 %
 10 Naum 2.2 2T                    : 2706   18  18   900    59.8 %   2637   40.6 %
 11 LoopMP 12.32 2CPU              : 2701   13  13  1680    56.9 %   2653   36.5 %
 12 HIARCS 11 MP UCI               : 2694   14  14  1480    58.1 %   2637   37.2 %
 13 Deep Shredder 10 UCI           : 2673   13  13  1980    52.0 %   2659   30.0 %
 14 Naum 2.1 MP                    : 2660   13  13  1740    48.9 %   2668   39.1 %
 15 HIARCS 11 UCI                  : 2651   15  15  1380    52.7 %   2632   35.7 %
 16 Fruit 2.3.1                    : 2646   18  18   900    51.4 %   2637   35.7 %
 17 Toga II 1.2.1a                 : 2637   13  13  1920    46.5 %   2661   34.4 %
 18 Deep Junior 10.1               : 2634   13  13  1980    46.3 %   2660   30.5 %
 19 Spike 1.2 Turin                : 2631   12  12  1920    45.6 %   2661   37.0 %
 20 Hiarcs X54 UCI                 : 2627   16  16  1140    51.2 %   2618   35.7 %
 21 Hiarcs X50 UCI                 : 2625   16  16  1140    51.0 %   2618   36.7 %
 22 Fruit 2.2.1                    : 2621   13  13  1920    44.1 %   2662   34.1 %
 23 Fritz 9                        : 2621   15  15  1380    48.1 %   2634   29.9 %
 24 Deep Sjeng 2.7 2T              : 2615   28  28   420    47.4 %   2633   30.0 %
 25 Glaurung 1.2 SMP               : 2582   13  13  1920    38.5 %   2663   27.9 %
 26 Chess Tiger 2007 UCI           : 2557   13  13  1980    35.3 %   2662   32.3 %
 27 Naum 2.0                       : 2546   16  16  1200    38.9 %   2624   36.2 %
 28 Scorpio 1.91 2CPU              : 2544   14  14  1680    31.8 %   2676   31.0 %
 29 Chess Tiger 15.0               : 2523   17  17  1140    35.8 %   2624   34.5 %
 30 Deep Pharaon 3.5.1             : 2522   13  13  1980    30.8 %   2663   29.3 %
 31 Deep Frenzee 3.0               : 2502   14  14  1980    28.2 %   2664   24.1 %
 32 Scorpio 1.8 2CPU               : 2501   16  16  1380    31.1 %   2639   30.0 %

[PauloSoare]

Preserve analyses and next best are very good for
analyses, and 100 elo points in a top program !!!

I am confused or Alex said that he didn´t developed Naum
after Naum 2.1?

Paulo Soares

[naum]

Hi,

In order to cut some of the suspense here is a little info about the new Naum:

- 100 ELO is just an estimate. It will be less against some opponents and more against others.
Since everyone is interested about the score against Rybka232a, the best score Naum got so far is +18 -25 =37 (tested on AMD3000 1CPU 32-bit no EGTBs).
Problem is that the version that does best vs Rybka isn't the strongest against other opponents.

- Many have noticed 'less speculative'. It's been mentioned by Shredder and Fritz too. That's one of the main things you need to do if you want to be competitive vs Rybka.
What I noticed is that even a very small increase in my king safety parameters will cause score vs Rybka to drop like a stone. This is one of the main 'secrets' of Rybka. Rybka is an expert in punishing any inaccuracies especially in the area of king safety.
So if you want to give a high king safety eval (like Naum 2.2 would often do), you better be damn sure you evaluated it correctly.
Less speculative should also apply to your search algorithm, but I don't think this is anything new. Fruit team have already mentioned and implemented that.

- Material imbalance improvements added only around 30 ELO. I don't know if this is the main weapon of Rybka. In my oppinion it's the 'less speculative' part I mentioned above.

- Playing style will hopefully be more like Rybka. That was one of the goals. If you can't beat them, join them But this doesn't mean Naum will play exactly the same as Rybka, and for those who want more aggresive playing style, I will probably provide an option for that.
I am fully prepared for the clone accusations. They already started and I didn't even release the engine.
My answer to all of them is 'I really don't care'. If you think it's a clone then don't buy it and leave me alone. I am not going to read or answer any of them.

- One of the reason for the new price is that I got more negative feedback than positive regarding my pricing.
I am surprised. I hoped to influence some to lower their astronomical prices, but it seems prices have gone up, not down.
Funny thing is that the only reasonably priced engine on the market is Rybka.

- And about the release date. I am not done yet. Not all features are implemented and there are still some possible strength improvements that I want to try. The main drag is the testing since it may take up to 1 week to properly test some of the features (and no, I don't need beta testers, I prefer to do my testing in house where I have a full control).
I am in no hurry, because as soon as I release the engine it will become the prime target for tuning of Rybka3.
Smart thing might be to wait until Rybka3 is released, but then Rybka3.1 would soon come etc. etc.

Regards,
Alex

[Alessandro Scotti]

Hi Alex,
that's very good news for computer chess, I hope you manage to squeeze some other elo from this new version so it will be even more fun when it comes out!

[Tony Thomas]

- Playing style will hopefully be more like Rybka. That was one of the goals. If you can't beat them, join them But this doesn't mean Naum will play exactly the same as Rybka, and for those who want more aggresive playing style, I will probably provide an option for that.
I am fully prepared for the clone accusations. They already started and I didn't even release the engine.

If you are talking about one of my posts, then it was just a joke. I had one of my previous post deleted (were I said it was a clone), I was trying to get people to try a new engine.

[Uri Blass]

1)I wonder if Naum earns more than rybka from time.

I think it may be interested to test naum against rybka at unequal time control.

If for example naum can score 50% against rybka at 2 minutes for naum against 1 minutes of rybka and can also score 50% against rybka at 5 minutes for naum against 3 minutes for rybka and can also score 50% against rybka at 40 minutes for naum against 30 minutes for rybka then it can be a good reason for correspondence players to hope that naum is better at long time control.

I think that an interesting test should decide about target of 50% score and simply give the time handicap that naum needs to get the relevant 50%.

2)I wonder what time control was used for your +18 -25 =37 result against rybka.

Uri

[Tony Thomas]

Funny thing is that the only reasonably priced engine on the market is Rybka.

Zappa is still reasonably priced. So was the previous versions of Naum.

[naum]

Hi Alessandro,

There are 2 possibilities:
1. I hit the sweet spot with the current version and every subsequent change will cause Naum's score vs Rybka to drop significantly.
2. I have at least 4-5 features from Naum 2.2 that I still didn't try in the new version. They bring at least 50 ELO in Naum 2.2. If they prove good in Naum3 things could get interesting

But really, I don't want to start a race with Rybka and it's army of IMs, GMs, testers and fanatic followers. If I start feeling any pressure, this would stop being fun and I would soon quit.

Regards,
Alex