A data science team is planning to train a series of models using XGBoost version 1.2 or newer on Amazon SageMaker for a predictive analytics project. Given the computational requirements of these models, which instance type would be the most appropriate for optimizing training performance and cost-efficiency on SageMaker?