GPU-accelerated instances for big data processing, scientific computing, and machine learning inference.