AWS Inferentia vs AWS Trainium2

The Trainium2 delivers 10.4x the FP16 throughput of the Inferentia (667 vs 64 TFLOPS dense). The Trainium2 also carries 88 GB more memory (96 GB vs 8 GB).

Inferentia: NeuronCore-v1, 2018 Trainium2: NeuronCore-v3, 2024

FP16 dense lead

10.4x

Trainium2: 667 vs 64 TFLOPS

Memory

8 vs 96 GB

88 GB more for the Trainium2

Bandwidth

50 GB/s vs 2.9 TB/s

Trainium2 moves data faster

Download comparison sheet Printable version →

AWS Inferentia

Full specs →

What is the AWS Inferentia?

The Inferentia is an ASIC from AWS on the NeuronCore-v1 architecture, an accelerator built for AI workloads. Inferentia full specifications →

Can I rent the Inferentia in the cloud?

Flopper is not tracking any live rental listings for the Inferentia yet. The provider directory lists the clouds that rent comparable accelerators on demand. GPU cloud providers →

How does the Inferentia compare with other accelerators?

This page sets the Inferentia against the Trainium2. The Inferentia is the older generation of the two. Both are AWS parts. The compare tool builds the same comparison for any two parts in the catalogue. All GPU comparisons →

Where do the Inferentia figures come from?

Every Inferentia figure on this page traces to AWS's own datasheet or specification table. How Flopper verifies specifications →

AWS Trainium2

Leads on FP16 Full specs →

What is the AWS Trainium2?

The Trainium2 is an ASIC from AWS on the NeuronCore-v3 architecture, an accelerator built for AI workloads. Trainium2 full specifications →

Can I rent the Trainium2 in the cloud?

Flopper is not tracking any live rental listings for the Trainium2 yet. The provider directory lists the clouds that rent comparable accelerators on demand. GPU cloud providers →

How does the Trainium2 compare with other accelerators?

This page sets the Trainium2 against the Inferentia. The Trainium2 is the newer generation of the two. Both are AWS parts. Flopper also compares it head to head with the H100 SXM5 80GB, across every published precision, memory and power. Trainium2 vs H100 SXM5 80GB →

Where do the Trainium2 figures come from?

Every Trainium2 figure on this page traces to AWS's own datasheet or specification table. How Flopper verifies specifications →

Specifications

Inferentia

Trainium2

Architecture

NeuronCore-v1NeuronCore-v3

Launch Year

20182024

Form Factor

——

Memory

8 GB96 GB

Memory Bandwidth

50 GB/s2.9 TB/s

TDP

——

Process Node

——

Spec

Inferentia

Trainium2

Architecture NeuronCore-v1NeuronCore-v3

Launch Year 20182024

Form Factor ——

Memory 8 GB96 GB

Memory Bandwidth 50 GB/s2.9 TB/s

TDP ——

Process Node ——

Performance (TFLOPS)

Inferentia

Trainium2

181 TFLOPS

667 TFLOPS

2,563 TFLOPS sparse

64 TFLOPS 667 TFLOPS

2,563 TFLOPS sparse

64 TFLOPS 667 TFLOPS

2,563 TFLOPS sparse

1,299 TFLOPS

2,563 TFLOPS sparse

FP6

128 TOPS

Precision

Inferentia

Trainium2

181 TFLOPS

667 TFLOPS

2,563 TFLOPS sparse

64 TFLOPS 667 TFLOPS

2,563 TFLOPS sparse

64 TFLOPS 667 TFLOPS

2,563 TFLOPS sparse

1,299 TFLOPS

2,563 TFLOPS sparse

FP6

128 TOPS

Figures are dense. Where a vendor publishes a figure it sits under the dense one and never decides a winner. Hover any precision for its definition, or read the GPU glossary.

FLOPS by Precision

What actually differs

The Trainium2 is the newer part: NeuronCore-v3 (2024) against the Inferentia's NeuronCore-v1 (2018).

The Trainium2 carries 88 GB more memory (96 GB against 8 GB), so larger models fit on one GPU before you shard. Bandwidth is 50 GB/s against 2.9 TB/s, which sets the ceiling on decode speed.

Weight the FP8 and FP16 rows for LLM work and the FP64 row for HPC.

New to the terms? The glossary explains FP32, TF32, BF16, FP16, FP8, INT8, Sparsity and TFLOPS.

Frequently asked questions

What is better, the Inferentia or the Trainium2?

The Trainium2, on throughput. At FP16 precision the Trainium2 reaches 667 TFLOPS dense against 64 TFLOPS for the Inferentia. The performance table on this page lists every published precision for both GPUs.

Which should you choose, the Inferentia or the Trainium2?

Choose the Trainium2 for training and high-throughput inference: it has 10.4x the FP16 throughput of the Inferentia. It also carries 88 GB more memory, so larger models fit on one GPU before you shard.

Which has more memory, the Inferentia or the Trainium2?

The Trainium2 carries 96 GB of memory versus 8 GB for the Inferentia. Memory bandwidth is 50 GB/s for the Inferentia and 2.9 TB/s for the Trainium2.

Get Comparison Updates

New GPUs added weekly. Be the first to see how they compare.

Open in Full Compare Tool → Inferentia full specs Trainium2 full specs Estimate cluster cost in the calculator →

AWS Inferentia vs AWS Trainium2

AWS Inferentia

What is the AWS Inferentia?

Can I rent the Inferentia in the cloud?

How does the Inferentia compare with other accelerators?

Where do the Inferentia figures come from?

AWS Trainium2

What is the AWS Trainium2?

Can I rent the Trainium2 in the cloud?

How does the Trainium2 compare with other accelerators?

Where do the Trainium2 figures come from?

Specifications

Performance (TFLOPS)

FLOPS by Precision

What actually differs

Frequently asked questions

What is better, the Inferentia or the Trainium2?

Which should you choose, the Inferentia or the Trainium2?

Which has more memory, the Inferentia or the Trainium2?

Related comparisons

Get Comparison Updates

Stay Updated on GPU Specs