| Industry | Universities |
| Location | Erlangen / Nuremberg |
| Company | Friedrich-Alexander University |
| Customer since | 02.2024 |
| Services | infra:u Private Cloud Stack |
How FAU Erlangen implements a GPU self-service with the infra:u Private Cloud Stack
Research and teaching today require immediate access to computing power. When students need to analyze large datasets or train complex AI models, they request flexible Kubernetes environments or dedicated high-end GPUs. The traditional IT request process simply blocks this highly dynamic environment.
This is exactly the challenge faced by the Erlangen National High Performance Computing Center (NHR@FAU) and the Regional Computing Center Erlangen (RRZE) of Friedrich-Alexander University (FAU) Erlangen-Nuremberg. The researchers' requirement was clearly formulated: They were looking for a streamlined cloud experience that provides the necessary resources at the push of a button, without long approval loops and preferably without IT intervention.
NHR@FAU and RRZE have excellent facilities and invest in state-of-the-art hardware. Outsourcing the compute-intensive workloads to global US hyperscalers was out of the question for the university from the outset: Location, devices, research data, and access must remain under the university's control.
However, building and especially operating a highly complex, scalable cloud platform including virtualization, Kubernetes as a Service (KaaS), and Ceph storage requires significant resources and specialized expertise that is needed elsewhere internally. The university was therefore looking for a way to retain physical control over the systems while outsourcing the operational base operation of the cloud platform.
The decision was made in favor of the "Software & Service" operating model with UPONU's infra:u Private Cloud Stack. This model allows for a clear division of roles: The university provides all physical hardware, from servers to NVIDIA Blackwell GPUs to network infrastructure.
In return, UPONU delivers the cloud layer, implements it on the 21 hypervisors at NHR@FAU and RRZE, and from that point on provides 3rd level support and assists with service.
“The stack runs on our own hardware in our own data center. We built the cloud layer in close collaboration with UPONU and successfully manage it together.”
Lukas Kraus, NHR@FAU – Systems & Services, FAU Erlangen-Nuremberg
In addition to physical data control, FAU also ensures complete independence at the software layer through this model. The infra:u stack is consistently based on open standards.
This protects the computing center from dependence on proprietary licensing models and unpredictable price dictates of individual virtualization providers such as VMware. The operational gain for the IT management lies in absolute data portability: Workloads can be migrated freely at any time. The university's IT budget thus flows into maximum computing power for researchers, rather than into blocking hypervisor licenses.
By outsourcing cloud operations, static hardware has been transformed into a fully managed, self-service cloud platform. Today, around 500 virtual machines run productively on the cluster at the Erlangen National High Performance Computing Center (NHR@FAU) and the Regional Computing Center Erlangen (RRZE).
Researchers, staff, and students have access to impressive performance data:
- 1,024 CPU threads on 21 hypervisors
- 22 terabytes of RAM
- 1.5 petabytes of Ceph storage
- 12 NVIDIA RTX 6000 Blackwell GPUs (96 GB VRAM each) for inference and AI workloads
In this setup, the Erlangen National High Performance Computing Center (NHR@FAU) and the Regional Computing Center Erlangen (RRZE) act as pure enablers for research. The usage concept is deliberately lean: The computing center provides the resources, while the students decide on the content themselves. If demand for resources continues to rise, the hardware can be seamlessly expanded by FAU in the same architecture, while UPONU supports the management of the cloud in close collaboration with FAU.
“We wanted an environment where someone can request a GPU and have the resources ready within minutes. The infra:u stack provides us with exactly this self-service for researchers, staff, and students.”
Lukas Kraus, NHR@FAU – Systems & Services, FAU Erlangen-Nuremberg
Are you ready for your sovereign cloud infrastructure?
Talk to our cloud architects and plan a solid, tailor-made solution on the infra:u stack.
