Bare Metal Server GPU

Benefit from the power of a fully dedicated physical server with our BMS GPU service, designed specifically for your most demanding artificial intelligence (AI) and machine learning (ML) needs.

Orange Business Bare Metal Server GPUs are designed for mission-critical workloads requiring isolation, performance and reliability. They meet the following requirements:

  • AI training and inference: adapted to SLM, LLM, fine-tuning and RAG (retrieval augmented generation) models.
  • Intensive data processing: for massive analysis in finance, healthcare, telecoms, etc.
  • Advanced simulation and visualization: useful in R&D for complex calculations.
  • Security and compliance: ideal for sensitive environments thanks to their hardware isolation.
  • Real-life development: test and validate applications in dedicated environments.
Dedicated to a clientEach Bare Metal Server GPU (BMS GPU) is reserved exclusively for a single customer, guaranteeing optimum performance and total confidentiality.
Without virtualization layerEliminate intermediate virtualization for a more direct computing experience and maximum utilization of hardware resources.
Installation of specific OS and applicationsPre-installation according to your specific needs for operating systems and GPU drivers.
Hosted and managed by Cloud AvenueOur expert team takes care of the complete hosting and management of your BMS GPU server, allowing you to focus on your core business without worrying about infrastructure. Choose Cloud Avenue’s BMS GPU service for a robust, flexible IT solution fully tailored to your specific requirements.
Server typeCPU & CoreRAM DDR5System diskNVMe storage GPUTotal VRAM
BMS Server GPU 4*H100 PCIe2x Intel Xeon Gold 6548Y, 32c @ 2.5 GHz1536 GB (24×64)2x 480 GB RAID 12×7.68 TB RAID 14x NVIDIA H100 NVL 94GB376 GB
BMS Server GPU 2*L40S PCIe2x Intel Xeon Gold 6548Y, 32c @ 2.5 GHz1024 GB (16×64)2x 480 GB RAID 12×7.68 TB RAID 12x NVIDIA L40s 48GB96 GB
BMS Server GPU 4* L40S PCIe2x Intel Xeon Gold 6548Y, 32c @ 2.5 GHz1024 GB (16×64)2x 480 GB RAID 12×7.68 TB RAID 14x NVIDIA L40s 48GB192 GB
BMS Server GPU CRAY 8*H100 SXM2x Intel Xeon Platinum 8452Y, 36c @ 2.2 GHz4096 GB (64×64)2x 960 GB RAID 12×7.68 TB RAID 18x NVIDIA H100 NVL 80GB (SXM)752 GB
Servers GPU 4*H100, 2*L40S et L40S

These models support GPU Direct Storage (GDS), enabling GPUs to access data stored on NVMe drives directly, significantly reducing data access latency and accelerating heavy-duty processing.

Server Cray

The CPUs offered in the Cray BMS are designed to excel in highly parallelized environments, thanks to their advanced architecture and larger cache.

The 8 H100 NVL 94 GB (SXM) graphics cards are linked by an inter-GPU SXM 5 bus (NVLink protocol) with a total bandwidth of 3.35 Tbps. This ensures ultra-fast communication between GPUs, essential for distributed training of large AI models, efficient parallelization of computations and reduced latency when processing massive workloads. The server is equipped with InfiniBand, an IB Monoport 400 Gbps MDR OSFP / GPU card.

Available operating systems

OS nameVersion
Ubuntu – LTS22.04, 24.04
RHEL – Red Hat Enterprise Linux8.6, 9.0, 9.1, 9.2
  • Ubuntu : NVIDIA drivers pre-installed according to compatibility.
  • RHEL (Red Hat) : NVIDIA drivers pre-installed according to compatibility.

The Bare Metal Server GPU service doesn’t include the operating system license. That means the Customer must provide the relevant license in BYOL mode (Bring Your Own License).

Cloud Avenue performs the initial installation of the operating system. The configuration, management, maintenance and updates are under the Customer’s responsibility after the server has been delivered.

Each BMS GPU server will be allocated local storage configured with 2 x 7.68 TB NVMe disks. This local SSD will be pre-installed with the operating system and can be used to store data.

The usable capacity for a client is divided by 2, due to RAID 1 mode replication.

As an option, 2×7.68 TB RAID 1 disks can be added, within the limits of the available GPU BMS slots.

Each BMS uses 2 x 25 Gbps LACP cards, providing access to a network of up to 50 Gbps.

By default, one VLAN is already configured, but it is possible to have up to 6 VLANs. The IP addressing plan is imposed and the network size is /24. The public IP is in the IaaS tenant, which means that the BMSs are not directly connected to the Internet, but only via the tenant (NAT).

The order form must specify which tenant and T1 the BMS is to be connected to.

Each BMS uses 2 x 25 Gbps LACP cards, providing access to a network of up to 50 Gbps.

Once the GPU BMS has been deployed, all essential information (such as configuration IP address, root password, etc.) is sent to the customer via the customer portal.

GPUs BMSs can be accessed by the customer in several ways:

  1. Bounce server: Customers can use an existing virtual machine already deployed in their client, or create a dedicated bounce server. They can then use a secure shell (SSH) or remote desktop (RDP) to access the GPU BMS.
  2. SSL VPN access
  3. VPN provided by T1: IPsec and L2VPN VPNs can be activated on Edge Gateway at no extra cost.
  • Local backup solution: Installed by the customer or provided by the application hosted on the server. The customer’s solution can use a dedicated space based on SharedNFS for storage.
  • Billing: GPU BMS is invoiced for a full month.

Cloud Avenue offers you a fixed monthly subscription per server (not variable by token or usage).

  • Cancellation: BMS GPU can be canceled at any time. Any month already started will be billed in full.

To order a GPU BMS server on Cloud Avenue, you need an existing tenant (Org VDC).

To order a BMS GPU server from Cloud Avenue, please get in touch with your regular sales contact, who will help you fill out a special BMS GPU order form.

Information to be submitted when ordering a GPU BMS

Organisation nameDisplayed on your contract
BMS GPU TypeGPU 4*H100, GPU 2 x L40s or GPU4 x L40s, 8H100 CRAY
OS NameUbuntu /RedHat and version
Service connectionGateway T0/T1

Once your order has been processed by our teams, you will be notified of the availability of your GPU BMS by e-mail.