Service Specification ccloud³ GPU (Cloud GPUs)

This English version is provided for convenience only. The German version is legally binding; in the event of any discrepancy between the two versions, the German version shall prevail.

1. Brief description

centron ccloud³ GPU (“Cloud GPU”) provides virtual machines (“VMs”). Each instance consists of a dedicated combination of virtual computing power (vCPU), graphics processing units (GPU), main memory (RAM) and local SSD storage. These resources are specifically optimised for compute-intensive workloads, in particular for:

• Artificial intelligence (AI) and machine learning (ML),
• The training and inference of neural networks,
• Complex data analyses and high-performance computing.

The Customer may freely choose the dimensioning of the resources within the system limits. The GPU VMs are provisioned, scaled and deleted by the Customer independently on a self-service basis via the customer portal or corresponding APIs.

2. Key features

ccloud³ enables virtual machines to be created, used and deleted according to individual requirements.

Depending on the selected model, one or more GPUs can be used within a virtual machine. High-performance GPU technologies are used for this purpose, which are particularly suitable for high-performance computing.

Use takes place on a self-service basis via the centron portal or via application programming interfaces (APIs).

3. Service plans

The service plans consist of various GPU-based instance types (“flavors”) that differ in terms of vCPU, RAM, and the number and type of GPUs.

The instances are available in different generations and include, in particular:

• Instances with NVIDIA RTX A4000 GPUs
• Instances with NVIDIA Quadro RTX 6000 GPUs
• Instances with NVIDIA A100 (40GB) or A100 (80GB) GPUs
• Instances with NVIDIA RTX 6000 Ada GPUs

The combination of CPU, RAM, SSD and GPU is indicated in the ordering process. All instances have local disks. Local disks constitute persistent storage.

A current overview of all available GPUs can be viewed via the centron portal or via the API.

4. Metrics

Billing is usage-based per virtual machine per unit of the respective resource on an hourly basis, with each hour or part thereof being charged in full. The following metrics are charged hourly per virtual machine:

• GPU: GPU / hour
• vCPU: vCore / hour
• RAM: GB / hour
• SSD: GB / hour

The billable period begins with the creation of the virtual machine and ends with its deletion. Periods during which a VM is operated in a stopped state without resource reservation (“shelving”) are also charged.

Additionally used resources, such as Object Storage or Backup Storage, are billed separately in accordance with the applicable Service Specifications.

5. SLA specifics

A monthly availability of 99.5 % applies to virtual machines with Cloud GPUs.

Virtual machines continue to be deemed available even if, due to incidents affecting other centron products or services, they temporarily have no access to their disks.

States caused by the Customer itself, in particular the stopping or shutting down of virtual machines, shall not be deemed unavailability.

6. Backup

Backup and restoration are not part of the services and are entirely the responsibility of the Customer.

This includes, in particular:

• the configuration of the virtual machines
• the data on local disks
• the data on attached storage (e.g. block storage)

Separate services or customer-side solutions are required to implement backup strategies.

7. Additional terms

The Customer is entitled to install and operate its own software on the virtual machines. Responsibility for licensing this software lies entirely with the Customer, unless expressly agreed otherwise.

Operating system images provided by centron may be used to create virtual machines. The portfolio of these images may be extended or adapted at any time without existing instances being affected.

By using these operating system images, the Customer accepts the respective licence terms of the manufacturers. A separate licence agreement is concluded between the Customer and the respective manufacturers.

The Provider assumes no warranty for selected operating system images and makes no availability commitments in respect of them. Likewise, the complete management of the operating system and of the software installed on it, including installation, maintenance, updates and security, lies exclusively with the Customer.

The Customer is responsible for the security of its virtual machines and of the applications operated on them.

ccloud³ uses GPUs from the manufacturer NVIDIA. The use of these GPUs and of the associated drivers is subject to NVIDIA’s licence terms as applicable from time to time. These apply in addition to the contractual agreements between the Provider and the Customer.

Where these licence terms contain restrictions on use, these also apply to the use of the Cloud GPUs.

The Provider is entitled to terminate the provision of the service with immediate effect if:

• NVIDIA’s licence terms are violated and the violation is not remedied,
• licence agreements between the Customer and NVIDIA cease to apply, or
• the necessary supply or licensing relationships with NVIDIA no longer exist.

The GPUs can only be used in conjunction with supported operating systems and drivers. The respective technical documentation must be observed in this regard.