In this study, we propose and evaluate a comprehensive power consumption model for GPU-based data centers that integrates the energy consumption of various task types, including language processing, speech recognition, and image generation, along with the power consumption of other components such as the CPU, memory, network, and storage. Using real GPU power consumption data, we measured the server energy consumption. To compensate for the limited experimental data and ensure accurate evaluation of average power consumption per unit time after model convergence, we utilize the TimeVAE model to generate supplementary data. The PCA and t-SNE visualizations validate the effectiveness of the TimeVAE model in preserving data distribution and structure. Our comprehensive model provides valuable insights into the energy consumption patterns in GPU-based data centers and offers practical guidance for optimizing energy usage to improve overall efficiency and sustainability.