Yes, it seems quieter without a current stated deadline. But I still wake up in the middle of the night remembering the old stated deadline. I had predicted April 10 as my most pessimistic date but that has been passed without a wimper. Now it's anyone's guess again. It could be June-July-August...

I think we should ask the creator of Zap that question, as he is presently closest to achieving that goal.  If I were guessing, if Rybka stood perfectly still and did not change there would be somebody right at her level in 6-12 months and somebody clearly surpassing in two years.  (This is just a speculative, meaningless guess, extrapolating from the gradual improvements we've seen over the past several years.)  My impression is that the area where the most substantial improvements might be made would be in coding for multiprocessors, when you consider that today's quads will be passe by 2010.  Have you guys seen Intel's press conference from late last month, when they laid out the CPU road-map through 2009?  Very impressive.  Imagine 32nm quads, fully integrated, with dynamic hyperthreading support...on a dual socket board 16 threads will not only be possible, but very much in play...clock speed something north of 4 GHz I would guess.

Actually, I think the major of people think the same way as you. You know the old saying that the " squeaky wheel gets the oil " , in this case its the those who are impetuous make them selves more apparent then most who are resolved to hang and wait the refinements to be made, and we will be there when it comes, " insha'Allah," and then only if, " ma sa Allah." Isn't that the way it goes or did I get that wrong.

Let us not tell about 2.3.1 but 2.3. Here was no time pressure but this version was a big disappointment for me. Bugs very rarely but simple quite weak performance after three month development. This can happen! But it isn´t very nice. 2.3.1 may be a progress when the bugs are away. But I´m not totally convinced. The main reason: 2.3.1 don´t change her mind; she is pigheaded. She looks very deep in the main variant and when there is no fail down she will play this although there are better moves. All versions of Rybka do this a little bit (may be the secret of Rybka?), but 2.3.1 overdo it, I believe. Perhaps on 64-bit Quads it´s not as bad as on 32-bit single core. For analysis, you always have to use the MV-modus.

Actually, with Rybka 1 and Rybka 2, the numbers are really "upside down".

The first versions (ie. Rybka 1.0 and Rybka 2.0) sound like major versions, but they were really raw. The final versions (1.2n and now 2.3.2) sound like afterthoughts, but these are essentially what the customers are paying for.

The numbers should be something like 0.79, 0.89, 0.90. 0.91, leading up to 1.0, and then again with 2.0.

Vas

Note that I read the following condition:
"I have the openings, which are problematic for chess programs (King’s Indian, Scandinavian, Evans gambit), deleted and replaced by more common ones, e.g. Grunfeld Indian, Dutch, Nimzo-Indian. Hereby I reached a relatively wide spectrum, namely per 4 games of the opening category A -E."

I think that this condition is not a good condition because if some engine has a relative advantage in the opening that you describe as problematic openings you cannot see it in the rating list.

The only reason not to include an opening is if the position is not balanced but if computers score is not balanced (say 80% for white) when the problem is not that the position is not balanced and computers can lose against humans even when they play white because humans know the right plan for black then it is better to include the opening in the rating list and an author who implements knowledge how to play with black should get an higher rating for his(her) program.

Uri

Hi Roland,

At CCRL we do have the freedom to run that matches that interest us individually however when you combine these you do get reliable ratings. I also run a huge number of gauntlets with common opponents that gives better accuracy.

As for rules we do have rules - we use common opening books for all pairings limited to a maximum of 12 moves 24 ply - the time controls are adjusted to a control machine etc etc.

Our rules can be seen here: http://kd.lab.nig.ac.jp/chess/discussion-board/viewtopic.php?t=1486

(For 40/4 the time control is 1/10th and we use less hash 64->128mb per thread.)

We also have a number of features that enable you to compare versions easily

From the index page you can select engines for comparison and get results like this:

http://computerchess.org.uk/ccrl/404/cgi/compare_engines.cgi?class=None&only_best_in_class=1&e=Rybka+2.2+64-bit+2CPU&e=Rybka+2.3.1+64-bit+2CPU&print=Rating+list&profile_step=50&profile_numbers=1&print=Score+with+common+opponents&table_size=100&match_length=30&sort_tables=by+rating

This shows Rybka 2.3.1 64-bit 2CPU and Rybka 2.2 64-bit 2CPU with ratings and common opponents - individual result should be viewed carefully as with a small number of games less than say 100 almost anything is possible.

We also have a 'pure list' - there is info on this on the page

http://computerchess.org.uk/ccrl/404/rating_list_pure.html

Unfortunately the pure list does require a very large number of games and opponent overlap for inclusion (which takes a while).

If you have any comments please register on our forum and post we are always happy to hear comments/suggestions as there is alway room for improvement.

All the best

Shaun