ASCEND 910B3 vs RTX 4090
The cheapest RTX 4090 ($0.16) costs 69% less per hour than the cheapest ASCEND 910B3 ($0.51).
Per unit of performance, the RTX 4090 is 55% cheaper: $0.25 versus $0.56 per A100-equivalent hour.
The ASCEND 910B3 has 64 GB of memory against 24 GB on the RTX 4090.
| Measure | ASCEND 910B3 | NVIDIA GeForce RTX 4090 |
|---|---|---|
| Lowest per GPU-hour | $0.51 | $0.16 |
| Median per GPU-hour | $0.51 | $0.58 |
| On-demand from | — | $0.19 |
| Spot from | — | $0.16 |
| Reserved from | $0.51 | $0.18 |
| Lowest per A100-equivalent hour | $0.56 | $0.25 |
| Live offers | 1 | 427 |
| Providers | 1 | 91 |
| Memory | 64 GB HBM2e | 24 GB GDDR6X |
| Memory bandwidth | 1,600 GB/s | 1,008 GB/s |
| FP16 / BF16 | 313 TFLOPS | 165.2 TFLOPS |
| FP8 | — | 330.3 TFLOPS |
| Performance vs A100 80GB | 0.91× | 0.63× |
| Architecture | 达芬奇 (Da Vinci) / 昇腾 910B | Ada Lovelace |
| Power | 350 W | 450 W |