FEN[226] = "6r1/p1pq1p1p/1p1p1Qnk/3PrR2/2n1P1PP/P1P5/4R3/6K1 w - - 0 1"; // f5h5 e5h5 g4g5 h5g5 h4g5 h6h5 e2h2 h5g4 h2g2 g4h5 f6f3 h5h4 f3g3 h4h5 g2h2 d7h3 g3h3 g6h4 h3h4 h5g6 h4h6Chessnut1071 wrote: ↑Sun Oct 01, 2023 12:08 amUmm, the way I process the reported numbers Myrddin is 790x faster than your program, in terms of nodes, and The King about 19,000x. Not 10x and 200x.hgm wrote: ↑Sat Sep 30, 2023 10:57 pm Umm, the way I process the reported numbers Myrddin is 790x faster than your program, in terms of nodes, and The King about 19,000x. Not 10x and 200x.
Note, however, that finding a mate and proving that it is the fastest mate is not the same thing. It might be for you, using a fixed-depth alpha-beta search, but these other programs of course use all kind of selective extensions and reductions. Otherwise it would of course be impossible to see mate-in-10 at 4 ply. It must have searched at least 19 ply on the branch that leads to the checkmate, otherwise it could not know it is there. But that leaves open the possibility that there would be a faster mate in the branches that were considered unpromising, and were only searched to 4 ply.
About this 'generic evaluation'... Fairy-Max virtually does not have any evaluation at all, other than piece values, a weak attraction to the center for P, N, B and K, and a penalty of half a Pawn for each King move in the branch leading up to the leaf where the evaluation is made. But it does have check extension, null move with R=2, and late-move reduction of 1 ply on all non-captures other than the hash move. So I think you are still on the wrong track by assuming this has anything to do with evaluation.
different FEN!
Using the FEN above, I tested the solution with and without the TT. Results as follows:
1) With TT: time to solution: 161.085 seconds; nodes searched = 310,019,642
2) Without TT: time to solution = 1244.05 seconds; nodes searched = 1,556,880,385
The TT table reduced time to solution by 80% and the number of nodes searched by 87%.
