Your team is in the process of developing a convolutional neural network (CNN)-based architecture from the ground up. Initial experiments conducted on your on-premises CPU-only infrastructure have shown promising results, but the model's convergence is slow. To expedite the model training process and shorten time-to-market, you are considering conducting experiments on Google Cloud virtual machines (VMs) equipped with more powerful hardware. It's important to note that your code doesn't involve manual device placement, and it hasn't been encapsulated within the Estimator model-level abstraction. Given this context, which environment should you choose for training your model?