Rent an L40 by the hour
On PowerPod an L40 costs from USD 0.80 an hour today, billed by the second, on one verified host. The price you see when you rent is locked for that instance.
What an L40 is good for
Datacenter cards from the Ada generation, strong at inference and at graphics-heavy AI work. With 45 GB of memory it is a good fit for:
- Serving 70B-class models with 4-bit weights (48 GB and up is comfortable)
- Fine-tuning 13B–14B models with LoRA, or 30B-class models with QLoRA
- SDXL and FLUX image generation and LoRA training at large batch sizes
- Long-context inference for 7B–14B models
Price over time
Where the hosts are
All L40 hosts listed today are in United States. Each offer shows its own measured reliability before you rent.
Typical machine
Median across the hosts listed today; each offer shows its exact machine before you rent.
| Memory bandwidth | 624 GB/s |
| Compute | 72.4 TFLOPS |
| Highest CUDA version | 12.8 |
| CPU cores for you | 24 |
| System memory for you | 35 GB |
| Download speed | 139 Mbps |
Questions
How much does it cost to rent an L40?
On PowerPod an L40 costs from USD 0.80 an hour today, billed by the second, on one verified host. The price you see when you rent is locked for that instance.
How much memory does the L40 have?
45 GB of GPU memory.
What can I run on an L40?
Serving 70B-class models with 4-bit weights (48 GB and up is comfortable). Fine-tuning 13B–14B models with LoRA, or 30B-class models with QLoRA. SDXL and FLUX image generation and LoRA training at large batch sizes. Long-context inference for 7B–14B models.
Can I pay in rupees?
Yes. Accounts in Nepal are billed in Nepali rupees and accounts in India in Indian rupees, paid with local methods; everyone else pays in US dollars. Prices can be shown in about 160 currencies.