gleperlier wrote:Seems Mark's solution leads to a draw like Qa3 or Qd4. But Qxe5 is more sexy
I don't think it draws. Have you taken a look at the lines from George on the Rybka forum? They are pretty convincing. I can't find a drawing line there.
Here is my best line:
1. Qxe5 fxe5 2. Rf1 a6 3. Bd1 Rc4 4. Bb3 b5 5. a4 Qe7 6. Kg2 *
(Alternately, 1. Qxe5 fxe5 2. Rf1 Rc7 3. Bd1 b5 4. Bb3 Rc4 - transposes into main line)
Arriving at this position:
[D]6kr/4q2p/p3P1pB/1p2p3/P1r1P3/1BP5/1P4KP/5R2 b - - 0 6
White wins here along any line I can see
6... g5 7. Rf5! Qg7 8. Bxg7 Kxg7 9. Bxc4 bxc4 10. Kf3 +-
6... Qe8 7. Rf3 g5 8. axb5 axb5 9. Ba2! Qe7 10. b4! Qg7 11. Bxg7 Kxg7 +-
6... Qd6 Rf7 7. Rf7 Qd8 8. axb5 axb5 9. Ba2! +-
JuLieN wrote:So was Marc's "solution" correct after all? Is Qxe5 the best move?
A very famous position, and, yes, it wins for white.
I doubt any engine can find Qxe5 on standard hardware, but Houdini shows that the position resulting from 24.Qxe5!!! fxe5 is won after 25.Rf1!
Quite amazing play by Gusev!
Here's Houdini's analysis using 3 threads on a Core i5-750, 2048 MB hash, using 5-man Scorpio bitbases (a new feature for the next Houdini release):
Houdini wrote:I doubt any engine can find Qxe5 on standard hardware, but Houdini shows that the position resulting from 24.Qxe5!!! fxe5 is won after 25.Rf1!
Very impressive that Houdini sees a win here!
25... Rc7 loses more quickly than 25... a6 I believe though. Does Houdini seen a better score than +2 after a6? Even along the line I posted earlier, it ends up in a rook endgame which was showing at about +1.
Houdini wrote:I doubt any engine can find Qxe5 on standard hardware, but Houdini shows that the position resulting from 24.Qxe5!!! fxe5 is won after 25.Rf1!
Very impressive that Houdini sees a win here!
25... Rc7 loses more quickly than 25... a6 I believe though. Does Houdini seen a better score than +2 after a6? Even along the line I posted earlier, it ends up in a rook endgame which was showing at about +1.
Aren't they just transpositions? Black has no useful moves, and probably has nothing else than a6, b5 and Rc4 when the white bishop goes to b3. White's plan is Bb3, a4, Rf3 , Kg2.
Houdini's main line also ends up in a rook ending which should be won for white.
Houdini wrote:...Here's Houdini's analysis using 3 threads on a Core i5-750, 2048 MB hash, using 5-man Scorpio bitbases (a new feature for the next Houdini release)...
Houdini wrote:...Here's Houdini's analysis using 3 threads on a Core i5-750, 2048 MB hash, using 5-man Scorpio bitbases (a new feature for the next Houdini release)...
What's the advantage for this type of EGTB ?
Scorpio bitbases can be completely loaded in memory when the engine starts - the Scorpio 5-man tables only require about 250 MB of RAM (peanuts on modern computers). That means that there are no slow disk access operations during the search, only memory operations.
The downside is that bitbases only contain Win/Draw/Loss information, there's no distinction between a position that is Mate in 1 and another position that is Mate in 100.
Houdini wrote:...Here's Houdini's analysis using 3 threads on a Core i5-750, 2048 MB hash, using 5-man Scorpio bitbases (a new feature for the next Houdini release)...
What's the advantage for this type of EGTB ?
Scorpio bitbases can be completely loaded in memory when the engine starts - the Scorpio 5-man tables only require about 250 MB of RAM (peanuts on modern computers). That means that there are no slow disk access operations during the search, only memory operations.
The downside is that bitbases only contain Win/Draw/Loss information, there's no distinction between a position that is Mate in 1 and another position that is Mate in 100.
Houdini wrote:A very famous position, and, yes, it wins for white.
I doubt any engine can find Qxe5 on standard hardware, but Houdini shows that the position resulting from 24.Qxe5!!! fxe5 is won after 25.Rf1!
Quite amazing play by Gusev!
It is a real difficult sacrifice. Probably it needs more than 50 depths and a very fast machine (8+ threads).
As far as I know engines have problems with the following similar and much easier queen sacrifice.
[d]rk6/p1p5/Bp2p3/1P6/q2B1P2/4K3/8/8 b - - 0 1
Houdini wrote:Aren't they just transpositions? Black has no useful moves, and probably has nothing else than a6, b5 and Rc4 when the white bishop goes to b3. White's plan is Bb3, a4, Rf3 , Kg2.
Houdini's main line also ends up in a rook ending which should be won for white.
Robert
Indeed, you are right, I missed that after Rc7, the PV had Rc4, so it ends up going down the same line. It's definitely a won rook endgame, but Houdini 1.5 was scoring it more around +0.8-1.