You work for a biotech startup that focuses on developing cutting-edge deep learning models based on the properties of biological organisms. Your team often engages in early-stage experimental phases with novel ML model architectures and frequently writes custom TensorFlow operations in C++. The models are trained on extensive datasets with substantial batch sizes, where a typical batch contains 1024 examples and each example is approximately 1 MB in size. The average size of an entire network, including all weights and embeddings, is around 20 GB. Considering these requirements, which hardware configuration would be the most suitable for your models?