INFERENTIA 2 vs L40S
The L40S has 48 GB of memory against 32 GB on the INFERENTIA 2.
| Measure | AWS Inferentia2 | NVIDIA L40S 48GB |
|---|---|---|
| Lowest per GPU-hour | — | $0.38 |
| Median per GPU-hour | — | $1.30 |
| On-demand from | — | $0.38 |
| Spot from | — | $0.45 |
| Reserved from | — | $0.49 |
| Lowest per A100-equivalent hour | — | $0.40 |
| Live offers | 0 | 284 |
| Providers | 0 | 85 |
| Memory | 32 GB HBM2e | 48 GB GDDR6 |
| Memory bandwidth | 820 GB/s | 864 GB/s |
| FP16 / BF16 | 190 TFLOPS | 362 TFLOPS |
| FP8 | 190 TFLOPS | 733 TFLOPS |
| Performance vs A100 80GB | 0.52× | 0.96× |
| Architecture | Inferentia2 (NeuronCore v2) | Ada Lovelace |
| Power | — | 350 W |
Where the offers sit
No live offers right now.