INFERENTIA 2 vs TRAINIUM 1
| Measure | AWS Inferentia2 | TRAINIUM 1 |
|---|---|---|
| Lowest per GPU-hour | — | $1.40 |
| Median per GPU-hour | — | $1.40 |
| On-demand from | — | $1.40 |
| Spot from | — | — |
| Reserved from | — | — |
| Lowest per A100-equivalent hour | — | $2.71 |
| Live offers | 0 | 1 |
| Providers | 0 | 1 |
| Memory | 32 GB HBM2e | 32 GB HBM2e |
| Memory bandwidth | 820 GB/s | 820 GB/s |
| FP16 / BF16 | 190 TFLOPS | 190 TFLOPS |
| FP8 | 190 TFLOPS | 190 TFLOPS |
| Performance vs A100 80GB | 0.52× | 0.52× |
| Architecture | Inferentia2 (NeuronCore v2) | Trainium (NeuronCore v2) |
| Power | — | — |
Where the offers sit
No live offers right now.