Your team is developing a convolutional neural network (CNN) architecture from scratch to address a specific problem domain. Initial trials on your current on-premises CPU-only infrastructure were promising, but the model's convergence is notably slow. To expedite the training process and shorten the time-to-market, your team is considering leveraging Google Cloud's powerful hardware offerings. Given that your code base lacks manual device placement and is not wrapped in an Estimator model-level abstraction, which Google Cloud environment should you select to train your model for optimal performance?