A10G vs L40S
The cheapest L40S ($0.38) costs 62% less per hour than the cheapest A10G ($1.00).
Per unit of performance, the L40S is 90% cheaper: $0.40 versus $4.02 per A100-equivalent hour.
The L40S has 48 GB of memory against 24 GB on the A10G.
| Measure | NVIDIA A10G 24GB (AWS) | NVIDIA L40S 48GB |
|---|---|---|
| Lowest per GPU-hour | $1.00 | $0.38 |
| Median per GPU-hour | $1.44 | $1.30 |
| On-demand from | $1.00 | $0.38 |
| Spot from | — | $0.45 |
| Reserved from | — | $0.49 |
| Lowest per A100-equivalent hour | $4.02 | $0.40 |
| Live offers | 22 | 284 |
| Providers | 3 | 85 |
| Memory | 24 GB GDDR6 | 48 GB GDDR6 |
| Memory bandwidth | 600 GB/s | 864 GB/s |
| FP16 / BF16 | 70 TFLOPS | 362 TFLOPS |
| FP8 | — | 733 TFLOPS |
| Performance vs A100 80GB | 0.25× | 0.96× |
| Architecture | Ampere | Ada Lovelace |
| Power | 150 W | 350 W |