NVIDIA L40S Pricing
Eli Mernit
Renting an NVIDIA L40S costs $0.38 to $7.58 per GPU-hour on demand across the providers below, though the top figure is an AWS VM that bundles 64 vCPUs and 512 GiB of RAM; the next-highest is $1.86. The getdeploying median is $1.54 across 34 providers. Beam's L40S is $0.76/hr with vCPU, RAM, and NVMe included.
Key takeaways
- Lium, a decentralized marketplace, has the cheapest verified in-stock L40S on getdeploying at $0.38/hr, half of Beam's price.
- Beam ($0.76/hr) and RunPod Community Cloud ($0.79/hr) come next, at about half the $1.54 median.
- AWS's $7.58 g6e.16xlarge has the same single GPU as its $1.86 g6e.xlarge, plus 60 more vCPUs and 480 GiB more RAM.
L40S cloud prices per hour, by provider
| Provider | SKU | On-demand $/GPU-hr | Notes |
|---|---|---|---|
| Lium | L40S PCIe | $0.38 | Decentralized marketplace; cheapest verified in-stock listing on getdeploying |
| Beam | L40S PCIe | $0.76 | On-demand machine; vCPU, RAM, and NVMe included |
| RunPod | L40S, Community Cloud | $0.79 | Secure Cloud $1.09 |
| Runcrate | L40S PCIe | $0.97 | Per-second billing; same listing on Sesterce |
| Verda | L40S 48GB | $1.54 | Spot $0.77 |
| AWS | g6e.xlarge | $1.86 | 4 vCPU, 32 GiB; us-east-1 |
| AWS | g6e.16xlarge | $7.58 | One GPU with 64 vCPU and 512 GiB; us-east-1 |
| Market median | — | $1.54 | Across 34 providers with an on-demand price (getdeploying) |
How these prices were collected
Prices are on-demand, per-GPU list prices in US dollars for single-GPU shapes, with spot excluded from the price column. AWS rates include each VM's vCPU and RAM. Marketplace data from Vast.ai, via getdeploying, supplies the spot and reserved floors.
- getdeploying's L40S price tracker for the Lium listing and the median
- RunPod's GPU pricing page for Community and Secure Cloud
- Runcrate's pricing page for its cheapest L40S offer
- Verda's pricing page for on-demand and spot rates
- AWS's EC2 On-Demand pricing page for g6e instances
L40S vs RTX 4090 / 5090: what the premium buys
Mostly memory: 48 GB, against 24 GB on the RTX 4090 and 32 GB on the 5090, in a passively cooled data-center card. On Beam the premium is small: $0.04/hr over a 5090 ($0.72) and $0.32 over a 4090 ($0.44). Market-wide it's wider, with getdeploying medians of $1.54 for the L40S, $0.64 for the 5090, and $0.47 for the 4090. Our RTX 5090 price comparison has the details.
If your model fits in 24 or 32 GB, the consumer cards cost less. On Beam they also run serverless, metered by the millisecond and scaling to zero: $0.69/hr for a 4090 or $1.09/hr for a 5090 under a committed-spend agreement, plus CPU and RAM. The L40S runs as an on-demand machine.
Where to rent an L40S for less
Marketplaces go lower. On getdeploying, an in-stock Lium L40S is listed at $0.38/hr, half of Beam's $0.76, and Vast.ai spot starts at $0.34/hr, with 6-month reservations from $0.49/hr. Marketplace machines come from individual hosts, so network, disk, and uptime vary by listing, and spot instances can be reclaimed.
RunPod Community Cloud costs $0.79/hr, three cents above Beam's machine price, which includes vCPU, RAM, and NVMe. RunPod's Secure Cloud tier costs $1.09/hr.
What this means for your workload
The L40S is an inference card. At two bytes per parameter, 48 GB holds about 24B parameters of FP16 weights before the KV cache, and FP8 or 4-bit quantization fits larger models. It suits serving an LLM behind an OpenAI-compatible API, batch inference, and image or video generation. It has no NVLink, so multi-GPU training belongs on A100 or H100 nodes.
FAQ
How much does an L40S cost per hour?
On-demand prices in the table run from $0.38 to $7.58 per GPU-hour, or $0.38 to $1.86 without AWS's CPU-heavy g6e.16xlarge. The getdeploying median across 34 providers is $1.54, and Beam's L40S machine is $0.76/hr.
What's the cheapest way to rent an L40S?
Marketplace offers are cheapest: Vast.ai spot from $0.34/hr and a Lium on-demand listing at $0.38/hr, both via getdeploying. After those, Beam at $0.76/hr and RunPod Community Cloud at $0.79/hr are the lowest on-demand rates in the table.
L40S vs RTX 5090: which should I rent?
Rent the 5090 if your model fits in 32 GB, since its median is $0.64 against $1.54 for the L40S. Rent the L40S when you need 48 GB on one card. On Beam, the extra 16 GB costs $0.04/hr on demand.
