Xonatron: Nigel, in your book you mention, "Months of analysis and more practical tests have thus far failed to establish the soundness of Sax's knight sacrifice. I did not believe it at the time but after errors by both sides in time pressure Sax finished the game rather nicely."I popped the position into Stockfish 6 (4096 MB hash + 8 cores), and it eventually sees the knight sacrifice as the best move.
...
Does not see it by depth 28. Notice the draw (0.00) evaluation:
28/41 00:23 184,379k 7,852k 0.00 bxa6 Rxa6 Nf5 Rxa3 bxa3 Nbd3 Bxd3 Nxd3 Re3 Qc5 Nxc4 Nxf2 Kxf2 Qxc4 Nxh6+ gxh6 Qg4+ Kh7 Qf5+ Kg8 Qg4+
First sees it at depth 29. Notice still close to draw (0.06):
29/41+ 00:36 287,587k 7,794k +0.06 Nxc4
Still at depth 29, it sees it as a distinct advantage (0.56):
29/48+ 01:07 535,811k 7,942k +0.56 Nxc4
Stopped at depth 34, the evaluation seems to stick (0.69):
34/56+ 08:11 3,748,930k 7,621k +0.69 Nxc4
...
(Note to chess engine connoisseurs that the move is detected immediately if the hash tables are already filled from previous moves. When you start the engine on this move, freshly, it takes much longer.)
Kibitzing
Annotations by 
sac: 20.Nxc4
