You are working on training a deep learning model using TensorFlow on a Google Cloud Platform (GCP) instance equipped with GPUs. The training data is distributed across multiple large files, and you've noticed that the GPU utilization is lower than expected due to the input pipeline's execution time. Your primary goal is to optimize the input pipeline to fully leverage the GPU's capabilities without increasing the overall cost. Which of the following strategies would be the most effective to achieve this goal? (Choose one)