Skip to main content

AWS Trainium 3 (Trn3)

Overview

AWS Trainium 3 reached GA on 2025-12-02 at re:Invent 2025, as the third-generation AWS custom AI training chip. Built on a 3nm process, it delivers 5.7 PFLOPS FP8 per chip (dense, officially 4.4× that of Trainium 2), 4× energy efficiency over Trainium 2, and 4× memory bandwidth improvement. Trn3 UltraServer connects 144 chips via NeuronLink (2.25× Trn2 UltraServer's 64 chips).

Trainium 3 is the centerpiece of AWS's "AI Factory" strategy — Amazon's internal Bedrock, Anthropic Claude, and AWS customer core training workloads are all migrating to Trn3.

Core Specifications

ItemSpecification
ArchitectureTrainium3 (NeuronCore-v4)
Process NodeTSMC 3nm
NeuronCore8 per chip (NeuronCore-v4)
HBM Capacity144 GB (estimated, 2× Trainium 2)
HBM Bandwidth~4.5 TB/s (estimated)
FP8 Compute (dense)5.7 PFLOPS
BF16/FP162,850 TFLOPS (estimated, half of FP8)
TDP~700 W
NeuronLinkNeuronLink-v4
Launch2025-12-02 GA (re:Invent 2025)

📌 Data convention: AWS Trainium uses dense compute as standard (consistent with AMD, Google); not directly comparable to NVIDIA sparse compute. 5.7 PFLOPS FP8 = dense (= 5,700 TFLOPS).

Trainium 2 vs Trainium 3 Upgrade Comparison

MetricTrainium 2Trainium 3Improvement
ProcessTSMC 4nmTSMC 3nm+1 generation
NeuronCore8 v38 v4Architecture upgrade
HBM Capacity96 GB144 GB (estimated)1.5×
HBM Bandwidth2.9 TB/s~4.5 TB/s~1.55×
FP8 Compute (dense)1,299 TFLOPS5.7 PFLOPS~4.4×
BF16/FP16667 TFLOPS2,850 TFLOPS (estimated)~4.3×
FP8 Compute per chip increase4.4×Official
Energy Efficiency (perf/watt)Official
Memory Bandwidth increaseOfficial
Launch2024-122025-12

⚠️ 5.7 PFLOPS = per chip (FP8 dense), while Trainium 2 is 1,299 TFLOPS/chip. Official 4.4× compute increase: 1,299 × 4.4 ≈ 5,716 TFLOPS ≈ 5.7 PFLOPS. Trust the official 4.4× compute improvement.

Trn3 UltraServer (Rack-Level)

ItemConfiguration
Chip Count144 Trainium 3 (4× Trn2 UltraServer's 64)
Total HBM~20.7 TB (144GB × 144)
NeuronLink-v4Fully connected, >10 TB/s bidirectional
FP8 Compute (rack)820 PFLOPS (dense, 144 × 5.7 PFLOPS)
BF16 Compute (rack)~410 PFLOPS
TDP (rack)~100 kW
Suitable Models400B+ parameter LLM training

Trn3 UltraServer = single rack can train 400B models. An EC2 UltraCluster (>10 racks) can support 1.4T+ parameter mega-model training.

Trn3 vs Trn2 UltraServer

MetricTrn2 UltraServerTrn3 UltraServerImprovement
Chip Count641442.25×
InterconnectNeuronLink-v3NeuronLink-v4New generation
Total HBM6.1 TB~20.7 TB3.4×
FP8 Compute~83 TFLOPS (64×1.3)~365 TFLOPS (144×2.54)~4.4×
Training Capability70B+ LLM400B+ LLM
Launch2024-122025-12

AWS Neuron SDK 3

  • Neuron SDK 3.x: PyTorch 2.4+ / JAX 0.4+ / TensorFlow 2.16+ fully optimized
  • Neuron Compiler 2.x: Auto compilation + graph optimization
  • NeuronX Distributed: Large-scale distributed training library (integrated with PyTorch FSDP)
  • NeuronX Nemo: LLM fine-tuning framework (Megatron-LM equivalent)
  • vLLM 0.7+ optimized: Low-latency inference

AWS Neuron = open-source ecosystem similar to ROCm, all SDKs are open source on GitHub (aws-neuron).

EC2 Instance Types

InstanceGPUConfigurationUse Case
trn3.48xlarge1 Trn3144GB HBMSingle-chip development
trn3.96xlarge2 Trn3288GB HBMSmall-scale training
trn3 UltraServer144 Trn320.7 TB HBMUltra-large-scale training

Pricing (Estimated)

InstanceHourly Price (on-demand)
trn3.48xlarge~$32 (estimated)
Trainium 2 equivalent~$16
Price increase
FP8 compute per dollar increase2.2× (based on 4.4× compute / 2× price)

AWS emphasizes: Trainium 3 is significantly better than NVIDIA H100 / H200 in FP8 compute per dollar (2-3×).

Use Cases

  • Ultra-large-scale LLM training (400B-1.4T models, UltraServer)
  • AWS Bedrock model pre-training (Anthropic Claude, Meta Llama, Mistral)
  • Cost-sensitive training (30-50% lower price than NVIDIA)
  • Energy-efficiency sensitive (4× perf/watt improvement)
  • ❌ Non-AWS deployments (Trainium only available on EC2)
  • ❌ Legacy NVIDIA ecosystem lock-in (high migration cost for CUDA-only code)

Vendor Information

ItemDetails
VendorAmazon Web Services (AWS)
First Release2025-12-02 (re:Invent 2025 GA)
Product Pagehttps://aws.amazon.com/machine-learning/trainium/
Cloud InstancesEC2 trn3.48xlarge / 96xlarge / UltraServer
SDKhttps://github.com/aws-neuron
PartnersAnthropic / Meta / Mistral / HuggingFace