Deep reinforcement learning (DRL) has garnered considerable attention and demonstrated remarkable efficacy in addressing complex decision-making problems across various domains. In the specific context of 5G and beyond networks, efficient resource allocation and management through task scheduling and placement are critical for meeting stringent quality-of-service requirements and accommodating diverse traffic demands. Nevertheless, existing literature often overlooks the associated costs of scheduling and placement decisions. This paper takes a step to fill this research gap by investigating the power consumption costs associated with the adoption of DRL techniques for task scheduling and placement problems in fog-cloud environments. To achieve this objective, several DRL models were implemented using established architectures from prior literature. A comprehensive evaluation was then conducted to compare the energy consumption and performance of these models, both during the training and execution phases, against typical models. The evaluation primarily focused on assessing the improvements in task response times achieved by the DRL models. The results of the study indicate that appropriately calibrated DRL models can indeed lead to enhanced task response times compared to the base models. However, it is worth noting that the decision-making process in the DRL models incurs significantly higher energy consumption, although this aspect remains inconsequential unless the models are deployed on battery-powered devices.