This was played in the last round of the Soviet Championships 1981 and Kasparov had to win to reach first. After losing this game Tukmakov ended up sharing fourth. Kasparov won 10 of his 17 games, and talks about the game here, he says it was a gamble, he considers black's opening line incorrect, but it worked:

Brilliant game by <Kasparov>, but the <Queen sac> is in no way a sacrifice.

It wins on the spot.

My <Shredder> just found the move at 1 second time control at 1 ply, if you can believe it.

No doubt I would have missed it OTB, however.

As Garry stated in video, yes, it is old Russian saying, attack someone 10 times in a row and they will blunder. Game was last round of Soviet Championship and Garry was 1/2 point behind Psakhis who won prior year. Kasparov needed win, played beloved KID and took risky chances. One of most interesting parts of video was when Jim pointed out 22...Rxa4 this move had never been considered by Garry (or anyone else) during the 19 years between game and video.

Houdini_20_x64: 29/68 55:22 37,498,314,849
0.00 22. ... Rxa4 23.b4 Ra1 24.Rc1 Rxc1 25.Qxc1
0.00 22. ... Re8 23.Qc1 Re4 24.b3 Bh3 25.gxh3
0.36 22. ... Bxb2 23.Rxd5 Qxc6 24.Rd6 Qc4 25.Bc3

Interestingly enough, at end of video for game (it has two parts, second part is deeper analysis based on discussion in first part). Garry realized that 22...Re8 had great psychological value and gave it exclam, much to Plaskett's distaste. Neither of them realized that, objectively, both moves scored dead even and, based on this alone, 22.Re8 was indeed by far the better choice, deserving exclam regardless of game outcome. Houdini engine was big help here! 23...Nb6 was riskier.

Houdini_20_x64: 30/64 14:40 10,554,302,106
+0.44 24.b3 Re2 25.Bc3 Bxc3 26.Nxc3 Re6 27.a5

As line suggests 25.Ba5 was small error. After 26.Qe7 position was equal, but 27.Nd4? gave nearly decisive edge to black.

Houdini_20_x64: 28/71 1:02:16 42,303,707,976
+2.05 27. ... Ra2 28.Bxb6 Bxe5 29.Qe1 Kh7 30.Rxe5