Machine Learning Acceleration Options

 

Name	Vendor	Cloud	Training	Inference
GPU-enabled server instances	NVIDIA	All	✅	✅
FPGA	Microsoft	Azure		✅
Neural Processing Unit (NPU)	Microsoft	Azure		✅
Tensor Processing Unit (TPU)	Google	Google Cloud	✅	✅
Inferentia	AWS	AWS		✅
Trainium (preview)	AWS	AWS	✅

 

Cloud vendors offer different options for accelerating machine learning, although some types of acceleration are better suited for training than inference, and vice versa. All cloud vendors support instance (VM) types with NVIDIA graphical processing units (GPUs). NVIDIA provides extensive library support for machine learning. (Amazon Web Services [AWS] and Microsoft also offer instance types with AMD Radeon GPUs; however, these target game streaming, rendering, and animation.)