B200 vs P40
The cheapest P40 ($0.034) costs 97% less per hour than the cheapest B200 ($0.99).
Per unit of performance, the B200 is 71% cheaper: $0.14 versus $0.49 per A100-equivalent hour.
The B200 has 180 GB of memory against 24 GB on the P40.
| Measure | NVIDIA B200 180GB SXM | NVIDIA Tesla P40 24GB |
|---|---|---|
| Lowest per GPU-hour | $0.99 | $0.034 |
| Median per GPU-hour | $7.89 | $0.21 |
| On-demand from | $0.99 | $0.034 |
| Spot from | $1.16 | — |
| Reserved from | $3.89 | $0.038 |
| Lowest per A100-equivalent hour | $0.14 | $0.49 |
| Live offers | 146 | 32 |
| Providers | 63 | 14 |
| Memory | 180 GB HBM3e | 24 GB GDDR5 |
| Memory bandwidth | 8,000 GB/s | 346 GB/s |
| FP16 / BF16 | 2,250 TFLOPS | — |
| FP8 | 4,500 TFLOPS | — |
| Performance vs A100 80GB | 6.96× | 0.07× |
| Architecture | Blackwell | Pascal |
| Power | 1,000 W | 250 W |