Amazon Web Services (AWS) is one of the most popular cloud platforms, and was made by Amazon.com. It provides many services, including cloud computing, storage, database services, content delivery, and other functionalities. In this section, we will only focus on virtual server services found on Amazon EC2. Amazon EC2 allows us to create multiple servers that can support the serving of our model and even the training routine. When it comes to serving the model for end users, you can read Chapter 9, Cruise Control - Automation, to learn about TensorFlow Serving. In training, Amazon EC2 has many instance types that we can use. We can use their CPU servers to run our web bot to collect data from the internet. There are several instance types that have multiple NVIDIA GPUs.

Amazon EC2 provides a wide selection of instance types to fit different use cases. The instance types are divided into five categories, as follows:

• General Purpose
• Compute Optimized
• Memory Optimized
• Storage Optimized
• Accelerated Computing Instances

The first four categories are best suited to running backend servers. The accelerated computing instances have multiple NVIDIA GPUs that can be used to serve models and train new models with high-end GPUs. There are three types of instances—P2, G2, and F1.