H100 SXM 80GB vs P40
The cheapest P40 ($0.034) costs 93% less per hour than the cheapest H100 SXM 80GB ($0.50).
Per unit of performance, the H100 SXM 80GB is 66% cheaper: $0.17 versus $0.49 per A100-equivalent hour.
The H100 SXM 80GB has 80 GB of memory against 24 GB on the P40.
| Measure | NVIDIA H100 80GB SXM5 | NVIDIA Tesla P40 24GB |
|---|---|---|
| Lowest per GPU-hour | $0.50 | $0.034 |
| Median per GPU-hour | $3.51 | $0.21 |
| On-demand from | $0.58 | $0.034 |
| Spot from | $0.50 | — |
| Reserved from | $1.41 | $0.038 |
| Lowest per A100-equivalent hour | $0.17 | $0.49 |
| Live offers | 317 | 32 |
| Providers | 83 | 14 |
| Memory | 80 GB HBM3 | 24 GB GDDR5 |
| Memory bandwidth | 3,350 GB/s | 346 GB/s |
| FP16 / BF16 | 989 TFLOPS | — |
| FP8 | 1,979 TFLOPS | — |
| Performance vs A100 80GB | 3× | 0.07× |
| Architecture | Hopper | Pascal |
| Power | 700 W | 250 W |