Abstract—In the rapidly evolving field of Mobile-Edge Computing (MEC), the demand for efficient Deep Reinforcement Learning (DRL) algorithms is critical due to the constraints of computational resources and the need for real-time processing. This paper introduces Optimized Nadam, an enhanced variant of the Nadam optimizer, specifically designed to address these challenges. By eliminating the computationally intensive product term, Optimized Nadam significantly reduces computational overhead while ensuring robust performance across varying load conditions. Experimental results demonstrate that Optimized Nadam achieves substantial reductions in Total Time Consumed, outperforming standard Nadam by up to 37.6% under typical load scenarios. Furthermore, the algorithm consistently exhibits a lower Average Time Per Channel, indicating superior convergence speed. Optimized Nadam maintains a high Normalized Computation Rate in dynamic environments with fluctuating load conditions, closely aligning with Nadam, thus showcasing its resilience and adaptability. The observed reduction in training loss across all test scenarios underscores Optimized Nadam’s efficiency in achieving rapid and stable convergence. These findings position Optimized Nadam as a viable alternative to traditional optimization algorithms such as Adam and Nadam, particularly in resource-constrained MEC deployments where computational efficiency and real-time processing are paramount.