ASCEND 310P vs ASCEND 910B4
The cheapest ASCEND 310P ($0.22) costs 28% less per hour than the cheapest ASCEND 910B4 ($0.31).
Per unit of performance, the ASCEND 910B4 is 65% cheaper: $0.48 versus $1.37 per A100-equivalent hour.
The ASCEND 910B4 has 32 GB of memory against 24 GB on the ASCEND 310P.
| Measure | ASCEND 310P | ASCEND 910B4 |
|---|---|---|
| Lowest per GPU-hour | $0.22 | $0.31 |
| Median per GPU-hour | $1.05 | $0.40 |
| On-demand from | $1.60 | — |
| Spot from | — | — |
| Reserved from | $0.22 | $0.31 |
| Lowest per A100-equivalent hour | $1.37 | $0.48 |
| Live offers | 7 | 2 |
| Providers | 2 | 2 |
| Memory | 24 GB LPDDR4X | 32 GB HBM2e |
| Memory bandwidth | 205 GB/s | 800 GB/s |
| FP16 / BF16 | 70 TFLOPS | 280 TFLOPS |
| FP8 | — | — |
| Performance vs A100 80GB | 0.16× | 0.65× |
| Architecture | 达芬奇 (Da Vinci) / 昇腾 310P | 达芬奇 (Da Vinci) / 昇腾 910B |
| Power | 72 W | 300 W |