"I already made one general change that would have avoided the 3 pawns up draw in Mexico"

Thanks Larry. That draw made me sad, it not a good way to draw when you are heavey material up. In general, do you think the change will avoid draws when it's a Bishops of opposite colours game? That would be very nice.

Overall, thanks for the change of the Mexico draw. Who knows the similar position might appear again in another match? :)

my opinion is, that it will be very hard to improve engines in the endgame so much, that they play them equally good as the middlegame (compared to humans). in the endgame you can't rely on search too much, because the themes like kingwalks and getting a specific piece coordination to start action and so on are very often too deep. but you also can't rely on general principles (which you might be able to teach by evaluation matters), because there are too many exceptions in endings as a rule. to me the only big jump in endgame strength could theoretically be achieved by inventing more abstract terms. such abstraction would be closer to a human way of solving endgame problems. the only good thing about this dificult task is, that it should be easier to do this abstraction in the endgame than in the middlegame as material is limited and ideas ar more concrete and easier to formulate.

I doubt that the majority of mainstream chess enthusiasts goes for top (fastest) performance for their chess engines. Also, I think that even among computer chess fans, there are many who are not willing to buy the latest fastest hardware (not even the latest fastest off-the-shelf hardware). For example, I am a computer chess freak since 20+ years, a chess player (hobby player) but I will not buy a quad soon. I also don't even have a Core Duo, only a D945. I also didn't buy MP Rybka but single Rybka and I am very satisfied with it.

I want to repeat a list of examples for using an engine in single core mode:

1. Kibitz 3 or 4 games simultanously from a GM tournament, online. Assign one kibitz engine each, on a quad.
2. Interactive analysis of a game, but user doesn't want to rely on just one engine. So he runs 2 or even 3, parallel.
3. Engine match with ponder=on on a dual core cpu. One core per engine, each.
4. Long analysis of a correspondence position in the background with one core, while the user can do other things on the same computer, without performance impact of one task on the other.

I think the relation of freestyle competition (and maybe, top level correspondence chess analysis and the like) to mainstream chess is like the relation of champ car, or formula one racing to normal car traffic. That's interesting and exiting, but it doesn't mean that anybody will buy a racing car and normal cars will disappear. :-)

Also, for occasional analyses, I can replace an additional investment of $ 1,000 ore more, by just letting calculate twice or three times as long. If I only need that analysis quality once in a while and not 24/7, than it is a simple way to save money. For the everyday analysis of my own patzer games, or for easy tactical analysis etc. single core is sufficient and economical.

     The version that played the final four games in Mexico already was much better in terms of scalability, and any further improvements in this category would probably only be worth a few Elo points on a quadcore, though perhaps on a 16 core machine there is still a decent amount left to gain. Anyway, if Vas can further improve scalab
ility without neglecting other important stuff I expect he will do so, but I don't think it's top priority anymore.

You have to notice that software is much more important than hardware (and OS)! Single core Shredder 11 is better than 4 core Shredder 10, single core Fritz 11 will be better than 4 core Deep Fritz 10. I demand, that Rybka 3.0 single is better than 4 core Rybka 2.3.2a. So if you are now in hurry, you can buy about 80-100 Elo by hardware (4 or 8 cores).
MP scaling is very impotant for offical tournaments like WCCC and freestyle, because here every Elo point counts. But for the normal customer of chess engines only the the playing strength and style of the single core engine counts. You have more than 1 core? Fine! But this isn´t really important for analysis of chess games.

My UNLEARNED guess says that it is going to be around 50 ELOs. I just don't like the 0 KN/s thing, I hope it is not going to be there in the next release. It seems as if Rybka stops when it gets to that 0 KN/s. I hope the Bishops of opposite colours will be taken care of as well. I hated that draw against Zappa when Rybka was a couple of pawns up, but a draw because of opposite colour Bishops.

That's a reasonable guess. But we may try out some fairly radical ideas. If they don't work out, we lose a few weeks work and the gain will be less, maybe only 35. If they do work out, we may have a pleasant surprise, with no way to predict what gains are possible.