CCRL Discussion Board

Here are two similar six men positions which may qualify as a test for KQkrp tablebases:

8/2p4P/8/kr6/6R1/8/8/1K6 w - - bm Kb1-a2; dm 72; pv 1.Ka2 Rh5 2.Rg5+!! Rxg5 3.h8Q; oo 7051
8/2p4P/6R1/kr6/8/8/8/1K6 w - - bm Kb1-a2; dm 72; pv 1.Ka2 Rh5 2.Rg5+!! Rxg5 3.h8Q; oo 7055

Ka2 is the only move to win.

After the rook sac, White queens the pawn and reaches a won five men ending, I hope DTZ<50.

The engines I tried are not able to discover the best move without the tablebases. But with the 5 man bases, they find the best move within 10-30s.

Regards,
Dhanish

an intersting test, so i took it a step further. with 4 man tbs, rybka 1.2 6 bit running on 3Gig athlon 64 using 768M of DDR-400 RAM found the solution to first in 6 secs using 64M RAM as a table cache. to second in 2 secs (!) both results #72 as yours. anyone want to try it with 3 man only? also, you need more variety than just a rook movement to create an 'extensive' test. shredder's endgame oracle would be fine handling this position with a wandering white rook almost immediately i would imagine, although that may be something that would require 6 man nalimov installed, so not sure...

Hi Gambit 3,

Actually, Rybka 1.0 Beta w32 finds Ka2 without any tablebases! In my tests, that was the only engine to do so.

Regards,
Dhanish

zappa mexico x64 found it in 32s. what engines were you trying?
rebel 12 still searching after 10 minutes.
rybka 1.0 beta 2s.
rybka 1.2 2s.

all engines with no tbs. to make sure i removed them from drive.
but taking a human look at the position Kc2 should also do it. after all, the rook sac is still on if you have k on a2 or on c2.
in any case, this seems at the very least to be very much in white's favour. (and last i checked i don't use tbs :> )

after a bit more analysis it seems the black rook is too able to check if white k goes other than a2. interesting. oh well...
BTW rebel took 296m53s to see Ka2 as >+6, but after 5 hours i'm not going to see how far it goes or how long it really takes.

oh and deep junior 9 took 143s (depth 34)