One hour
$0.79
1 GPU × 1h · compute only
Ada data-center GPU rental
Rent an NVIDIA L40S GPU for multimodal AI and rendering.
Direct answer
NVIDIA L40S combines 48 GB of GDDR6 ECC memory with Ada Tensor, RT and media engines. It fits production inference, fine-tuning, rendering and image or video pipelines that need more memory and data-center features than a GeForce GPU. GPURento lists L40S at $0.79 per GPU-hour in Paris and Frankfurt.
Reviewed by the GPURento infrastructure team · September 1, 2026
Cost scenarios
What one L40S costs at the listed rate.
One day
$18.96
1 GPU × 24h · compute only
Seven days
$132.72
1 GPU × 168h · compute only
Average month · 730h
$576.70
1 GPU × 730h · compute only
| Memory | 48 GB GDDR6 ECC | Official NVIDIA L40S configuration. |
|---|---|---|
| Architecture | Ada Lovelace | Tensor, RT and media acceleration in a data-center form factor. |
| On-demand rate | $0.79 / GPU-hour | Compute only before persistent storage. |
| Available regions | Paris · Frankfurt | Selectable through the funded dashboard. |
Best for
- Production LLM and vision inference that fits 48 GB
- LoRA and selected fine-tuning workloads
- Image, video and 3D rendering pipelines
- Teams needing ECC memory and data-center characteristics
Not the best fit when
- Very large models that require more than 48 GB on one GPU
- Training jobs that depend on NVLink
- Small jobs where a lower-cost RTX or A5000 is sufficient
Decision method
What to verify before you deploy capacity.
The content below separates published specifications, GPURento catalog references and decisions that still require a workload benchmark.
01
L40S is a mixed-workload GPU
The value of L40S is not one peak benchmark. It combines AI compute, 48 GB of ECC memory, ray-tracing hardware and three encode/decode engines. That makes it useful when one environment serves models and produces visual or video outputs.
02
Know the topology limit
NVIDIA lists L40S as PCIe Gen4 and does not list NVLink support. For tightly coupled multi-GPU training, compare H100, H200 or a cluster topology instead of assuming multiple L40S devices behave like one shared-memory accelerator.
- Measure the exact batch and concurrency
- Include media encode or decode in the benchmark
- Check whether the complete model and cache fit in 48 GB
03
Compare with RTX 5090 on system requirements
RTX 5090 can offer attractive single-GPU throughput with 32 GB. L40S adds 48 GB, ECC and data-center positioning. Choose from the memory ceiling, reliability requirements and end-to-end output cost rather than generation alone.
Catalog shortlist
Relevant GPU options.
AdaAvailable
NVIDIA L40S
48 GB GDDR6 ECC · $0.79/hr
Production inference, fine-tuning and video
EU West · ParisEU Central · Frankfurt
BlackwellAvailable
NVIDIA RTX 5090
32 GB GDDR7 · $0.69/hr
Fast single-GPU inference and media
EU West · ParisEU Central · Frankfurt
HopperAvailable
NVIDIA H100 SXM
80 GB HBM3 · $2.69/hr
Intensive training and FP8 inference
EU West · ParisEU Central · Frankfurt
Questions
Clear answers, including the limits.
How much does it cost to rent an L40S GPU?+
GPURento lists L40S at $0.79 per GPU-hour. One continuous 24-hour day is $18.96 before storage, and 730 hours is $576.70.
How much memory does NVIDIA L40S have?+
The L40S has 48 GB of GDDR6 memory with ECC and an official memory bandwidth of 864 GB/s.
L40S or RTX 5090?+
Choose L40S when 48 GB, ECC or data-center characteristics matter. RTX 5090 can be more attractive when the workload fits 32 GB and GeForce-class hardware is acceptable.
Does L40S support NVLink?+
NVIDIA does not list NVLink support for L40S. Multi-GPU workloads should account for PCIe and host networking rather than assume an NVLink topology.
Sources and scope
Last reviewed September 1, 2026. GPURento rates are current catalog references; provisioning state remains visible in the workspace, and manufacturer specifications do not substitute for workload benchmarks.
Continue the research
Deployment plan
Use L40S when one GPU must handle AI and media.
Create a funded workspace and select the L40S image, region, runtime and persistent storage you need.