AWS Trainium
Overview
Trainium is the first AWS chip built for training rather than inference, the part under EC2 Trn1 and Trn1n instances and the start of the line that runs through Trainium2 and Trainium3. Each chip is two NeuronCore-v2 cores over 32 GB of HBM at 820 GB/s, rated by AWS at 190 TFLOPS across FP16, BF16, its configurable cFP8 and TF32, 380 TOPS of INT8 and 47.5 TFLOPS of FP32, with 48 MiB of software-managed on-chip SRAM. It is the same two-core NeuronCore-v2 design as Inferentia2, which is why the two chips carry identical figures; what Trainium adds is the scale-out, with NeuronLink-v2 at 384 GB/s per chip joining sixteen chips in a 2D torus inside a trn1.32xlarge so their memory can be pooled, and stochastic rounding for training. AWS announced it at re:Invent 2020 and made Trn1 generally available in October 2022. It is sold only as EC2 capacity, and AWS publishes no power figure, process node or die size for it.
Performance
Peak theoretical throughput by precision type
| Precision | Peak |
|---|---|
FP64 | No verified data available |
FP32 32-bit floating point | 48TFLOPS |
TF32 TensorFloat-32 | 190TFLOPS |
BF16 Brain Float 16 | 190TFLOPS |
FP16 16-bit floating point | 190TFLOPS |
FP8 8-bit floating point | 190TFLOPS |
FP6 | No verified data available |
FP4 | No verified data available |
INT8 8-bit integer | 380TOPS |
Every figure here is dense. The vendor documents no structured (2:4) sparsity mode for this part, so there is no second number to quote.
Specifications
Architecture
NeuronCore-v2
Form Factor
No verified data available
Launch Year
2020
Process Node
No verified data available
Memory
32 GB HBM
Bandwidth
820 GB/s
TDP
No verified data available
Max power
No verified data available
Interconnect
384 GB/s NeuronLink-v2
direction and scope not stated by vendor
NeuronCore-v2
2
Spec Confidence
Official
Full Specifications
| Compute Engine | |
|---|---|
| NeuronCore-v2 | 2 |
| Memory | |
| Memory | 32 GB |
| Memory Type | HBM |
| Bandwidth | 820 GB/s |
| Interface Width | No verified data available |
| On-Die SRAM | 48 MB |
| Interconnect & I/O | |
| GPU-to-GPU | NeuronLink-v2 |
| Interconnect Bandwidth | 384 GB/s direction and scope not stated by vendor |
| Power & Thermal | |
| TDP | No verified data available |
| Max power | No verified data available |
| Enterprise Features | |
| Sparsity | No |
| General | |
| Form Factor | No verified data available |
| Architecture | NeuronCore-v2 |
| Process Node | No verified data available |
| Launch Year | 2020 |
Datasheet & Resources
Data Provenance
Every figure traced to a source
Primary Source
- Publisher
- AWS
- Published
- No verified data available
Data Quality
- Spec confidence
- Official
- Clock basis
- Boost
- Core precisions with figures
- 6 of 9
- Normalization
- All values in TFLOPS
Compare cloud providers offering on-demand GPU instances for AI training, inference, and HPC workloads.
Browse GPU Cloud ProvidersSimilar GPUs
Frequently Asked Questions
How many TFLOPS does the AWS Trainium have?
The AWS Trainium delivers 48 TFLOPS FP32, 190 TFLOPS FP16 and 190 TFLOPS FP8 at peak.
What is the power consumption of the AWS Trainium?
Flopper does not currently have a verified TDP (Thermal Design Power) figure for the AWS Trainium.
How much memory does the AWS Trainium have?
The AWS Trainium is equipped with 32 GB of memory with 820 GB/s of memory bandwidth.
What architecture is the AWS Trainium based on?
The AWS Trainium is based on the NeuronCore-v2 architecture, launched in 2020.
Stay Updated on GPU Releases
Get notified when new GPUs are added or specifications are updated.
No spam, unsubscribe anytime.