L40S vs P4
The cheapest P4 ($0.01) costs 97% less per hour than the cheapest L40S ($0.38).
Per unit of performance, the P4 is 20% cheaper: $0.32 versus $0.40 per A100-equivalent hour.
The L40S has 48 GB of memory against 8 GB on the P4.
| Measure | NVIDIA L40S 48GB | NVIDIA Tesla P4 8GB |
|---|---|---|
| Lowest per GPU-hour | $0.38 | $0.01 |
| Median per GPU-hour | $1.30 | $0.86 |
| On-demand from | $0.38 | $0.01 |
| Spot from | $0.45 | — |
| Reserved from | $0.49 | $0.34 |
| Lowest per A100-equivalent hour | $0.40 | $0.32 |
| Live offers | 284 | 46 |
| Providers | 85 | 7 |
| Memory | 48 GB GDDR6 | 8 GB GDDR5 |
| Memory bandwidth | 864 GB/s | 192 GB/s |
| FP16 / BF16 | 362 TFLOPS | — |
| FP8 | 733 TFLOPS | — |
| Performance vs A100 80GB | 0.96× | 0.03× |
| Architecture | Ada Lovelace | Pascal |
| Power | 350 W | 75 W |