Once a computer gets to an advantage of +/-1.0 or so, the win is virtually assured in all positions, barring a few weird theoretical draws and misevaluations of the score of the position (e.g. closed positions with no way to make progress), although the latter problem has gotten a lot better with NN engines.

(I used to play computer correspondence chess.)

Edit: I should add that +/-1.0 is roughly a pawn's worth of material. So a computer up a knight or bishop -- nominally 3 points of material -- should virtually always win.