Amazon data center Available

AWS Trainium

Trainium ยท Trainium 1st Gen Architecture

Enquire Now
AWS Trainium is Amazon's first-generation custom training chip, purpose-built for deep learning workloads on AWS. Each Trn1 node offers up to 512GB of HBM memory and 190 TFLOPS of BF16/FP16 performance. Trainium delivers up to 50% cost savings over comparable GPU instances for training tasks, making it ideal for customers looking to train models at scale with better economics.

Key Features

2 NeuronCores per chip 512GB HBM per node BF16/FP16 training EC2 Trn1 instances AWS Neuron SDK support

Full Specifications

Compute

Architecture Trainium
Process Node 7nm
NeuronCores 2
FP16 Performance 190 TFLOPS
BF16 Performance 190 TFLOPS

Memory

Memory Size 512 GB
Memory Type HBM (per node)
Memory Bandwidth 820 GB/s

Power & Physical

Form Factor Custom ASIC

Features & Connectivity

NVLink Support No
Multi-GPU Support Yes

Availability

MSRP (USD) Contact for pricing
Release Date Oct 2022
Status Available

Industries

AI/ML Cloud Computing Research

Use Cases

Deep Learning Training NLP Model Training Computer Vision Training Recommender Systems

Interested in the AWS Trainium?

Get pricing, availability, and bulk discount information from our team.

Enquire Now

Related GPUs

Amazon data center

AWS Trainium2
Available View Specs
Amazon data center

AWS Inferentia2

Memory

32GB HBM

FP16

190 TFLOPS

Available View Specs
Amazon data center

AWS Inferentia

Memory

8GB DDR4

Available View Specs