Retireborn: This may be the first game in which 7...c5 was played; in later years it appears frequently in the games of Sosonko, who used it as a solid if somewhat drawish weapon.
Not a lot happens until Torre decides to snatch a pawn with 31.Nxa7. He could have played the solid 31.Nd4 Qb7 32.Nf3 Bxf3 33.gxf3 Qxf3 34.Qxa7 with equality.
After the reply 31...Qd3! he finds he's actually going to be a pawn down, as he cannot defend e3 (32.Qb6 Qe2) or f4 (33.g3 Qf3!)
The resulting endgame is looked at rather briefly by Averbakh in his BvN book (1976 translation.) He thinks that White's 55.h4 is good enough to draw; one variation he gives is 55.h4 Kg6 56.hxg5 Kxg5 57.Ng1 h5 58.Nh3+ Kf5 59.Ng1 Ke5 60.Ne2 Ke4 61.Kxg2 Ke3+ 62.Kf1 Ba6 63.Ke1(!) Obviously Black can't take the knight now, but 63...h4 still wins, according to Tablebase.
Nevertheless, Black does let the win slip with 59...Kf4; although it's a very natural move, 59...h5 is the only move to win here, according Tablebase.
Torre still has to find two only moves 60.Nd3+ and 64.Ne3 to hold on.
Could Black have won the game without entering this difficult endagame? I asked Houdini and it came up with 47...Qe1+ (instead of 47...f3) 48.Kh2 h5 49.Nd6 Qg3+ 50.Kh1 f3 51.Qc8+ Bg8 52.Nf7+ Kh7 53.Qf5+ g6 54.Qxf3 Qe1+ 55.Kh2 Qe7 and the knight is trapped.
Very elegant, but perhaps difficult for practical players (now or then) to think like this....