Cloud-based radio access networks (C-RAN) are expected to face important challenges in the forthcoming fifth generation (5G) communication systems. For this reason, more flexible C-RAN architectures have recently been proposed in the literature, where the radio communication stack is partitioned and placed across different RAN nodes to tackle the 5G capacity and latency requirements. In this paper, we show that this functional split also supports energy efficiency, especially when it is combined with bandwidth adaptation. To this aim, we have built a dynamic hotspot prototype, where the hardware-accelerated physical-layer is placed in the remote radio head, and higher software-based layers are placed in a server (either directly connected or remotely accessible). This setup allowed us to experimentally evaluate the power consumption of key hardware modules when adapting the bandwidth and the modulation and coding scheme. The real-time operation of the testbed allows further experimentation with different 5G use cases and the evaluation of other key performance indicators. TRUE pub