Side by side datacenter GPU specifications from the Flopper.io database.
| AWS Inferentia2 | AWS Trainium | |
|---|---|---|
| FP32 TFLOPS | 47.5 | 47.5 |
| Memory | 32 GB | 32 GB |
| Memory Bandwidth | 820 GB/s | 820 GB/s |
| TDP | — | — |
| Precision | AWS Inferentia2 peak TFLOPS (dense) | AWS Trainium peak TFLOPS (dense) |
|---|---|---|
| FP64 | — | — |
| FP32 | 47.5 | 47.5 |
| TF32 | 190.0 | 190.0 |
| BF16 | 190.0 | 190.0 |
| FP16 | 190.0 | 190.0 |
| FP8 | 190.0 | 190.0 |
| FP6 | — | — |
| FP4 | — | — |
| INT8 | 380.0 | 380.0 |
| AWS Inferentia2 | AWS Trainium | |
|---|---|---|
| Architecture | NeuronCore-v2 | NeuronCore-v2 |
| Process Node | — | — |
| Launch Year | 2022 | 2020 |
| Form Factor | — | — |
| Memory Type | HBM | HBM |
| Interface Width | — | — |
| Interconnect | NeuronLink-v2 | NeuronLink-v2 |
| TDP | — | — |
| FP32 Efficiency | — | — |