ASCEND 910B vs L40S
The cheapest L40S ($0.38) costs 1% less per hour than the cheapest ASCEND 910B ($0.38).
Per unit of performance, the ASCEND 910B is 4% cheaper: $0.38 versus $0.40 per A100-equivalent hour.
The ASCEND 910B has 64 GB of memory against 48 GB on the L40S.
| Measure | ASCEND 910B | NVIDIA L40S 48GB |
|---|---|---|
| Lowest per GPU-hour | $0.38 | $0.38 |
| Median per GPU-hour | $0.63 | $1.30 |
| On-demand from | $0.52 | $0.38 |
| Spot from | — | $0.45 |
| Reserved from | $0.38 | $0.49 |
| Lowest per A100-equivalent hour | $0.38 | $0.40 |
| Live offers | 19 | 284 |
| Providers | 8 | 85 |
| Memory | 64 GB HBM2e | 48 GB GDDR6 |
| Memory bandwidth | 1,600 GB/s | 864 GB/s |
| FP16 / BF16 | 376 TFLOPS | 362 TFLOPS |
| FP8 | — | 733 TFLOPS |
| Performance vs A100 80GB | 1.02× | 0.96× |
| Architecture | 达芬奇 (Da Vinci) / 昇腾 910B | Ada Lovelace |
| Power | 400 W | 350 W |