No NVLink
L40S cards talk to each other over PCIe only. Splitting one model across two or four of them works but is slower than on NVLink-connected GPUs, so choose a model size that fits on one card if you can.
GPU rental prices · Ada Lovelace
NVIDIA L40S rental prices
NVIDIA L40S rental starts at $1.09 per GPU-hour (Runpod, PCIe, 48 GB) across the 5 offers we track from 4 providers. The typical price is $1.67, and one GPU running all month costs about $796 at the lowest rate. A 48 GB card with FP8 support but no NVLink, priced well below the H100. It suits inference of models up to about 30B at 8-bit, and image or video generation.
- Lowest per GPU-hour
- $1.09 Runpod
- Typical per GPU-hour
- $1.67 median, USD
- One GPU all month
- $796 at the lowest rate
- Providers
- 4 5 offers
On-demand prices
NVIDIA L40S price per GPU-hour
Every L40S offer in our catalog, cheapest first. Multi-GPU machines are divided by their GPU count, so you can compare one GPU with one GPU. Monthly cost assumes the GPU runs all 730 hours.
| Provider and offer | Version | GPU memory | GPUs | Per GPU-hour | Per month, 1 GPU | Checked |
|---|---|---|---|---|---|---|
| RunpodRunpod Pod · NVIDIA L40S | PCIe, 48 GB | 48 GB | 1 | $1.09 | $796 | 3 Nov 2026 |
| DigitalOceanL40S GPU Droplet · 1 GPU | PCIe, 48 GB | 48 GB | 1 | $1.57 | $1,146 | 3 Nov 2026 |
| VultrVultr NVIDIA L40S · 1 GPU | PCIe, 48 GB | 48 GB | 1 | $1.67 | $1,220 | 3 Nov 2026 |
| OVHcloudOVHcloud US l40s-180 | PCIe, 48 GB | 48 GB | 2$3.600/hr total | $1.80 | $1,314 | 3 Nov 2026 |
| OVHcloudOVHcloud US l40s-90 | PCIe, 48 GB | 48 GB | 1 | $1.80 | $1,314 | 3 Nov 2026 |
On-demand list prices before tax, from each provider's pricing page. Community and reserved prices, where a provider offers them, are lower; see the notes below. Full offer details, storage and billing terms.
Before you rent
What changes the price
Memory bandwidth sets the speed
At 864 GB/s the L40S has about a quarter of the H100 SXM's 3.35 TB/s. Chat models read the whole model from memory for every token, so expect slower replies than on an H100 even when the model fits.
Community prices
Runpod's Community Cloud lists the L40S at $0.79 an hour, against $1.09 on its Secure Cloud.
AI models
Which models fit on the L40S
Memory to load each model size on 48 GB GPUs, with a 20% allowance for runtime buffers, and what that many GPUs cost per hour at the lowest rate above. Long contexts and many parallel users need more memory. More than one GPU only works if your inference server splits the model across them.
| Model size | 4-bit | 8-bit | 16-bit |
|---|---|---|---|
| 8BLlama 3.1 8B, Qwen3 8B | 4.5 GB1 GPU · $1.09/hr | 9.0 GB1 GPU · $1.09/hr | 17.9 GB1 GPU · $1.09/hr |
| 32BQwen3 32B, Gemma 4 31B | 17.9 GB1 GPU · $1.09/hr | 35.8 GB1 GPU · $1.09/hr | 71.6 GB2 GPUs · $2.18/hr |
| 70BLlama 3.3 70B, Qwen 72B | 39.2 GB1 GPU · $1.09/hr | 78.3 GB2 GPUs · $2.18/hr | 156.5 GB4 GPUs · $4.36/hr |
| 120Bgpt-oss-120b class | 67.1 GB2 GPUs · $2.18/hr | 134.2 GB3 GPUs · $3.27/hr | 268.3 GB6 GPUs · $6.54/hr |
| 685BDeepSeek-V3.2 | 382.8 GB8 GPUs · $8.72/hr | 765.6 GB16 GPUs · more than one 8-GPU server | 1,531.1 GB32 GPUs · more than one 8-GPU server |
Weights only, from the parameter count. Check a specific model, context length and user count in the model hardware calculator.
Hardware
NVIDIA L40S specifications
| Specification | L40S |
|---|---|
| GPU memory | 48 GB GDDR6 with ECC |
| Memory bandwidth | 864 GB/s |
| FP8 Tensor | 733 TFLOPS (1,466 sparse) |
| Max power | 350 W |
| Form factor | PCIe, dual-slot |
| NVLink | No |
From NVIDIA's product pages. Tensor figures marked “sparse” assume structured sparsity; dense throughput is half.
Questions
L40S rental FAQ
How much does it cost to rent an L40S?
$1.09 per GPU-hour is the lowest on-demand price we track, from Runpod. The typical price is $1.67, and one GPU running all month costs about $796 at the lowest rate.
Can an L40S run a 70B model?
At 4-bit the weights need about 39.2 GB with a 20% allowance, which fits on one 48 GB card with under 9 GB left for the conversation cache. At 8-bit a 70B model needs 78.3 GB, or two cards.
Should I rent an L40S or an A100?
The L40S has 48 GB and FP8 support. The 80 GB A100 has more memory and about 2.2 times the bandwidth (1,935 GB/s on PCIe against 864 GB/s) but no FP8. For models over about 30B, or for long contexts, the A100's extra memory matters more.
Compare
Other GPU rental prices
Prices and availability change, and we don't check stock. Confirm the current price and region with the provider before you rent. Sources: www.nvidia.com, www.runpod.io, www.digitalocean.com, www.vultr.com, us.ovhcloud.com . See our affiliate disclosure.