Choosing Servers Algorithm Cloud supports three main types of servers:

  • Two GPU servers are supported: P2 and G3s. Both perform about the same but your choice should depend on the model size being used with the GPU.

  • Elastic Inferance: Add GPU acceleration to any Amazon EC2 instance for faster inference at much lower cost (up to 75% savings) compared to using a dedicated GPU machine. The downside is all that is supported is Tensorflow, and MXNet for model deployment. The GPU's described are attached to the standard servers attached for a paring of a partial GPU with the server described.
  • See Amazon's website for more information on how to deploy and information: Amazon Elastic Inference

  • CPU based servers with different amounts of CPU cores and Ram. Typically, choose 1 core 2GiB t2.small as it works well for most algorithms, ML and DL models where GPU acceleration is not required.
  • Two types of servers are supported, t and c instances. T servers can vary in network speed and compute ability so if you require very consistent inference algorithm speed then choose C instances at a higher cost.