Deep learning has evolved into the most important supporting technology for artificial intelligence (AI) and has achieved widespread application across various fields. However, the energy expenditure associated with training deep learning models has become increasingly significant, now representing an undeniable part of global carbon emissions. This chapter mainly focuses on techniques for achieving energy-efficient deep learning training. It first addresses the context of the significant energy consumption associated with training AI models. Then, it specifically focuses on optimization algorithms and parallel training methods—two key technologies for improving the efficiency of deep learning training. Following that, it presents additional supporting technologies that enhance the training efficiency of AI models. Finally, it provides an overview of specific strategies from a macro perspective.