Side by side datacenter GPU specifications from the Flopper.io database.
| AWS Inferentia | AWS Trainium3 | |
|---|---|---|
| FP32 TFLOPS | — | 183.0 |
| Memory | 8 GB | 144 GB |
| Memory Bandwidth | 50 GB/s | 4.9 TB/s |
| TDP | — | — |
| Precision | AWS Inferentia peak TFLOPS (dense) | AWS Trainium3 peak TFLOPS (dense) |
|---|---|---|
| FP64 | — | — |
| FP32 | — | 183.0 |
| TF32 | — | 671.0 |
| BF16 | 64.0 | 671.0 |
| FP16 | 64.0 | 671.0 |
| FP8 | — | 2517.0 |
| FP6 | — | — |
| FP4 | — | 2517.0 |
| INT8 | 128.0 | — |
| AWS Inferentia | AWS Trainium3 | |
|---|---|---|
| Architecture | NeuronCore-v1 | NeuronCore-v4 |
| Process Node | — | 3nm |
| Launch Year | 2018 | 2025 |
| Form Factor | — | — |
| Memory Type | DDR4 | HBM3e |
| Interface Width | — | — |
| Interconnect | NeuronLink-v1 | NeuronLink-v4 |
| TDP | — | — |
| FP32 Efficiency | — | — |