This document is relevant for: Inf1, Inf2, Trn1, Trn1n

Neuron Architecture#

The Neuron Architecture provides insights into Neuron enabled instances system, software and chip capabilities. The EC2 Trn and Inf instance architecture provides an overview of the EC2 instances powered by the Inferentia and Trainium chips (Neuron Devices), and the corresponding system features like inbox and network connectivity, memory hierarchy, and NeuronCores versions and capabilities. The Neuron model architecture fit provides insights to what is the best match between deep-learning model architectures and the NeuronCore version.

Trn and Inf instances#

EC2 Trn1/Trn1n Architecture
EC2 Inf2 Architecture
EC2 Inf1 Architecture

Trainium and Inferentia devices#

AWS Trainium Architecture
AWS Inferentia2 Architecture
AWS Inferentia Architecture

NeuronCores#

NeuronCore-v1
NeuronCore-v2

Neuron Model Architecture#

Neuron Model Architecture Fit Guidelines

Other#

Neuron Glossary

This document is relevant for: Inf1, Inf2, Trn1, Trn1n