L40S GPU Instances

Why choose NVIDIA L40S GPUs?

Powerful

Up to 1.2 times faster than the L4, suited for AI inference workloads.

High performance

Ada Lovelace architecture and 4^th generation Tensor Cores bring a noticeable performance boost to GenAI, image streaming, and 3D rendering.

Optimised

Optimised for virtualisation and GPU rendering (vGPU): ideal for 3D studios, engineering, and hybrid cloud environments.

Sovereign

Available in our Public Cloud, ensuring flexibility, transparency, and compliance with European standards.

Optimised for your AI and graphics workloads

AI inference and multimodal GenAI

Generate images, videos, and audio faster with 4^th generation Ada Lovelace Tensor Cores, built for streaming and GenAI.

3D rendering and visualisation

Get excellent rendering performance for CAD, design, simulation, or real-time creation tasks.

Hybrid AI + graphics workloads

The L40S provides the power needed to combine AI inference and GPU rendering within the same cloud environment.

Specifications

Technical specifications

GPU

1–4 GPUs per instance

GPU memory

48 GB per GPU

High-performance storage

NVMe storage on most instances

Public and private network

Up to 25 Gbps included

Automation

Management via your Control Panel, API, OVHcloud CLI, etc.

Secure and private

ISO27001, SOC certifications, health data hosting.

Our Cloud GPU range

H200

Up to 1.4 times faster than the H100. Ideal for LLM 65B+.

H100

AI benchmark for training and large-scale generative AI.

A100

Best balance of performance, cost, and AI flexibility.

V100

Reliable GPU for machine learning and scientific computing.

V100S

An upgraded V100 with higher bandwidth and frequency.

L4

Efficient and cost-effective for AI inference and video processing.

A10

AI + graphics versatility makes it ideal for inference and computer vision.

RTX 5000 Quadro

Designed for 3D rendering, visualisation, and professional design.

Ready to accelerate your AI projects?

Create an account and launch your services in minutes

Get US$ 200 in free credit to launch your first Public Cloud project

Maximise your ROI with flexible GPU infrastructure

Pricing transparency

Get great value for your money when creating images, videos, or AI models — with no billing surprises.

AI and graphical versatility

Leverage a single GPU for your generative AI, 3D rendering, and AI inference workloads. This provides you with unmatched flexibility for your production pipelines.

Sovereignty and compliance

Your data is hosted in a secure, transparent, and compliant European cloud, certified to meet regulations (GDPR, ISO, HDS).

Energy efficiency

Reduce your operating costs with an energy-efficient Ada Lovelace architecture — without compromises on performance.

How do I choose my GPU for inference?

Lightweight inference

L4: Budget-friendly and energy-efficient.

Medium-scale inference

L40S: more memory, improved performance, greater versatility.

Large models

A100/H100/H200 according to your bandwidth and performance needs.

Configure your GPU instances

Choosing your GPU for LLM inference

Discover how to set up your GPU architecture to meet the requirements of AI models.

Explore the key differences between our AI Notebooks, AI Training, and AI Deploy solutions

Choose the one that best suits your needs using the comparison table.

Deploy a GPU application on the Managed Kubernetes Service

We will assist you in setting up your GPU on Managed Kubernetes via the OVHcloud Control Panel and Helm.

Your questions answered

What service level agreement is guaranteed by OVHcloud on a GPU instance?

The SLA guarantees 99.99% monthly availability on GPU instances. For further information, please refer to the General Terms of Service.

Which hypervisor is used for instance virtualisation?

Just like other instances, GPU instances are virtualised by the KVM hypervisor in the Linux kernel.

What is PCI Passthrough?

Cards with GPUs are served via the physical server’s PCI bus. PCI Passthrough is a hypervisor feature that allows you to dedicate hardware to a virtual machine by giving direct access to the PCI bus, without going through virtualisation.

Can I resize a Cloud GPU instance?

Yes, Cloud GPU instances can be upgraded to a higher model after a reboot. However, they cannot be downgraded to a lower model.

Do GPU instances have anti-DDoS protection?

Yes, our anti-DDoS protection is included with all OVHcloud solutions at no extra cost.

Can I switch to hourly billing from an instance that is billed monthly?

If you have monthly billing set up, you cannot switch to hourly billing. Before you launch an instance, please take care to select the billing method that is best suited to your project.

What is a Cloud GPU?

A Cloud GPU is a cloud computing service that provides graphic processing units (GPUs) for tasks that require high computing power. Examples of these are graphic rendering, machine learning, data analysis, and scientific simulations. Unlike on-premises GPUs, which require a significant investment in hardware, cloud GPUs are more flexible and easier to scale. Users can access high-performance computing resources on demand, and only pay for what they use.

What is an H100 and A100 server?

Servers that are equipped with NVIDIA H100 and A100 GPUs are purpose-built to offer exceptional performance in HPC, AI, and data analytics.

What is NGC?

NVIDIA GPU Cloud (NGC) is a cloud computing platform offered by NVIDIA. It provides a comprehensive selection of software that is optimised for GPU acceleration in artificial intelligence (AI), machine learning (ML), and high-performance computing (HPC). NGC simplifies and speeds up the deployment of AI and scientific computing applications. It does this by providing containers, pre-trained models, SDKs, and other tools that are optimised to leverage NVIDIA GPUs.

Why use a Cloud GPU?

There are several advantages to using a Cloud GPU, especially for companies, R&D, and development teams in demanding fields, such as artificial intelligence (AI), graphics rendering, machine learning (ML), and high-performance computing (HPC).