First Layer Versus Last Layer

A 3-layer network with one unit per layer and a smooth activation $\phi$:

$a_0 = x, \qquad z_l = w_l\,a_{l-1}, \qquad a_l = \phi(z_l)$

On one batch, with input $x = 2$, the forward pass recorded these values:

and the loss contributes

$\frac{\partial L}{\partial a_3} = 2$

Compute the gradient at the first layer's weight, $\partial L / \partial w_1$, and compare its size with the gradient at the last layer's weight, $\partial L / \partial w_3$.

Which pair is correct?