How Many Steps Until It Has Converged?

A model is trained on the loss

$L(w) = 5(w - 2)^2$

with the update rule

$w \leftarrow w - \alpha\, L'(w), \qquad \alpha = 0.03$

starting from $w_0 = 10$.

The team does not watch the loss curve by eye. They use a fixed stopping rule:

Training is declared converged the first time the loss drops below $0.5\%$ of its value at the starting point.

How many steps are needed before the stopping rule fires?

Give a whole number of steps.