Domain names illustration

NVIDIA Rubin NVL8 GPU instances


NVIDIA Rubin NVL8

The Rubin generation is coming to the OVHcloud Public Cloud

The next step in computing power

Rubin succeeds Blackwell. With the HGX Rubin NVL8 platform, your AI and scientific computing workloads benefit from faster memory, next-generation interconnection and much more power.
  • 130 TB/s GPU bandwidth

    HBM4 memory bandwidth on the instance: data reaches the GPUs faster, and your calculations no longer wait.
  • 400 PFLOPS*

    In NVFP4 for AI inference, and 265 TFLOPS with double precision (FP64) for simulation: an instance for AI and scientific computing.
    *With sparsity. NVIDIA HGX Rubin NVL8 platform figures published by NVIDIA.
  • NVLink 6

    The 8 Rubin GPUs are connected via NVLink Switch (24 TB/s): the instance works like a single large accelerator.
  • Keep control

    Hosted on our European Public Cloud, with transparent pricing and no hidden fees.

One instance, three workload families

  • AI Training

    Agents and reasoning

    Agents complete series of steps, reasoning models produce many tokens. With 2 TB for conversation context and NVLink to distribute the model, serve more agents in parallel, even during peak hours.
  • AI models

    Serving and fine-tuning large models

    MoE models, long contexts, fine-tuning and post-training: 270 PFLOPS in NVFP4 to adapt recent open models to your data.
  • Start SQL Database

    Accelerated data analysis

    ETL, queries on large dataframes, vector indexing, embeddings for RAG: move processing that saturates your CPUs to GPU.

SPECIFICATIONS

Technical specifications

  • GPU

    GPU

    8 x NVIDIA HGX Rubin NVL8 GPUs
  • NVFP4 inference computing

    NVFP4 precision with sparsity supports inference workloads requiring high and consistent throughput.
  • NVFP4 post-training

    The announced power reaches 270 PFLOPS with sparsity, within a future European GPU infrastructure.
  • Automation

    Management via your Control Panel, API, OVHcloud CLI, etc.
  • Shield Cloud

    Security and compliance

    ISO27001, SOC certifications, health data hosting.

Ready to get started?

Create an account and launch your services in minutes

Benefit from $270 free on your first Public Cloud project

Frequently asked questions about NVIDIA Rubin NVL8

When will Rubin NVL8 instances be available?

The instances are expected to be available on this platform in 2027. You can submit your requirements to anticipate capacity.

What is NVIDIA Rubin NVL8?

Rubin is the generation of NVIDIA GPU that succeeds Blackwell. The HGX Rubin NVL8 platform brings together 8 Rubin GPUs in a single server, connected by the 6th generation of NVLink, with 2 TB of HBM4 memory.

Is Rubin NVL8 suitable for scientific computing?

Yes. With 265 TFLOPS in double precision (FP64) per instance and a memory bandwidth of 130 TB/s, Rubin NVL8 is suitable for simulations, molecular dynamics, or financial calculations that can be stored on a single server, as well as pipelines that combine simulation and AI.

What is the difference compared to the B300?

The B300 (Blackwell Ultra architecture) offers the largest memory capacity in our range, up to 2.3 TB per instance, and it's available now. Rubin NVL8 focuses on speed: faster HBM4 memory, a next-generation NVLink interconnect, and more computing power, for both AI and simulation.

Should I wait for Rubin or start on B300?

If your project needs to start now, the B300 is available today, with no commitment or with a 1-year or 3-year commitment. Your tools, APIs, and images remain the same: you will then be able to switch to Rubin NVL8 without changing cloud.

What will the configurations be and how much will it cost?

Configurations (vCore, RAM, storage), locations, and pricing will be communicated at launch. Contact us to discuss.