We study the problem of energy efficiency maximization (EEmax) with joint beamforming and subarray selection, by taking into account the non-linear power amplifier (PA) efficiency in a multi-user multiple-input single-output system. The subarray selection problem is formulated using the concept of perspective formulation with additional penalty term in the objective function. To tackle the resulting challenging mixed-Boolean non-convex optimization problem, we rely on continuous relaxation and successive convex approximation framework where a convex problem is solved in each iteration. Numerical results demonstrate the achieved energy efficiency gains of the subarray selection and show that non-linear PA efficiency has a significant impact on the optimization. We also observe that on contrast to using linear PA efficiency model, the non-linear PA efficiency model yields the fact that it is better to stay silent rather than transmit with very low transmit power.