RF Power Amplifiers (PA) correspond to the most energy-consuming elements in the transmission chain. From one side, the communication community studies MIMO channels and multiuser precoding using simplified models for the power amplifier: the PA is assumed to operate in the linear region, where DC consumption is proportional to output power plus a constant. Conversely, in the RF design community, simple tone signals are commonly used for testing, where energy efficiency optimization aims to increase drain efficiency. The result is high drain efficiency for test signals; however, communication efficiency is not considered. Since the optimization is done at the circuit level, no channel considerations are done in the case of beamforming. Similarly, neither signal quality improvement due to precoding nor the impact of inter-user interference over the MIMO channel is considered. Furthermore, we want to analyze the impact of Digital Predistortion (DPD) on the system efficiency. Hence, PA models must consider non-linear variations in DC consumption in low input power and overdrive regimes. For improving energy efficiency, a wide range of effects must be considered (precoding algorithms, PA non-linear models, including linearity and DC consumption, DPD), with different objective functions among the two communities. This paper aims to bridge the two worlds: using a non-linear PA model, which includes DC consumption and non-linear effects, and multiuser precoding algorithms, the relation between estimated rate and DC energy consumption is studied, focusing on the impact of DPD on signal quality and energy efficiency. The linear part of the signal is separated from non-linear distortion terms and interference using the MIMO Bussgang decomposition, which has shown to be an excellent tool for explaining the origins of rate degradation. The results show a trade-off between maximum energy efficiency and maximum achievable rate. It is shown that DPD can improve the maximum achievable rate if inter-user interference is correctly managed via precoding; nevertheless, it cannot improve the signal quality if interference among users dominates the total system rate.