AWS Trainium

Inferentia/Trainium Architecture

Active

Launched November 2021

Core Specifications

VendorAWS
ArchitectureInferentia/Trainium
Form Factor
VRAM32 GB
Memory Bandwidth
TDP200 W

Compute Performance

PrecisionTFLOPs
BF16190

Performance Benchmarks

llm train

ConfigurationPrecisionPerformanceSource
EC2 p5.48xlarge pricing (on-demand)13.5 cost_per_hourView
LLaMA 70B, BF16, optimized for AWS4,200 tokens_per_secondView

Quick Stats

Peak Performance
190
TFLOPs (BF16)
Efficiency
0.95
TFLOPs per Watt

Similar XPUs

View other AWS GPUs or compare across vendors