H20 vs RTX 5090
The cheapest RTX 5090 ($0.22) costs 57% less per hour than the cheapest H20 ($0.51).
Per unit of performance, the RTX 5090 is 52% cheaper: $0.24 versus $0.49 per A100-equivalent hour.
The H20 has 96 GB of memory against 32 GB on the RTX 5090.
| Measure | NVIDIA H20 96GB | NVIDIA GeForce RTX 5090 |
|---|---|---|
| Lowest per GPU-hour | $0.51 | $0.22 |
| Median per GPU-hour | $1.06 | $0.87 |
| On-demand from | $1.02 | $0.34 |
| Spot from | $0.51 | $0.25 |
| Reserved from | $0.71 | $0.22 |
| Lowest per A100-equivalent hour | $0.49 | $0.24 |
| Live offers | 26 | 405 |
| Providers | 11 | 84 |
| Memory | 96 GB HBM3 | 32 GB GDDR7 |
| Memory bandwidth | 4,000 GB/s | 1,792 GB/s |
| FP16 / BF16 | 148 TFLOPS | 209.5 TFLOPS |
| FP8 | 296 TFLOPS | 419 TFLOPS |
| Performance vs A100 80GB | 1.03× | 0.92× |
| Architecture | Hopper | Blackwell |
| Power | 400 W | 575 W |