DrMAL: Thanx <qqdos> for pointing out game. Once again in Velimirovic attack, there is option for white to either play positional game with 14.b4 or to continue attack with key sac 14.Nf5 and then there is option for black to accept sac (14...exf5) or decline (14...b4 if black already went b5). Here are quick evals before and after 14.Nf5, character of lines are much more important than scores.Houdini_20_x64: 27/66 15:05 9,243,789,640
0.00 14.b4 Ne5 15.f4 Nc4 16.Kb1 Bd7 17.Bc1 0-0
-0.08 14.Nf5 exf5 15.Nd5 Qd8 16.exf5 Bb7 17.f6
Houdini_20_x64: 28/71 32:19 21,171,347,564
+0.04 14. ... exf5 15.Nd5 Qd8 16.exf5 Bb7 17.f6
-0.25 14. ... b4 15.Nxe7 bxc3 16.Nxc8 Rxc8 17.bxc3
In game, Balashov declined sac with 14...b4 counterplay move instead, I think this gives much better chances for black. Position is still sharp but not nearly as much as accepting sac which IMO is very dangerous and should be avoided unless very familiar with theory lines and their move orders. Computer often fools average player this way, engines presently do not give any indication of line character or of consequences to common mistakes.
Few moves later Balashov took sharper option of taking pawn 16...cxb2+ before taking N this does not transpose with 16...Rxc8 because of Q position.
Houdini_20_x64: 27/66 10:34 6,744,740,920
-0.29 16. ... Rxc8 17.bxc3 0-0 18.Rd4 Qxc3 19.Kb1
-0.37 16. ... cxb2+ 17.Kxb2 Rxc8 18.Rd4 0-0 19.Ra1
Play was accurate on both sides. White chose 19.Rhd1 (19.Ra1 computes as best) also good move. With 20.Kb1 instead of 20.Qd3, it seemed clear that critical part of game was over, players were headed for draw, very instructive Velimirovic attack game.
