Computing infrastructure

The AI Service Centre offers a high-performance computing (HPC) infrastructure for training and developing AI models that is unique in Hesse. In addition to infinitely scalable computing power of up to 632 A100 GPUs, the computing cluster offers additional non-mainstream hardware for the research and development of specialised AI solutions. For example, 4 Graphcore bow-200 nodes and an Nvidia Developer Toolkit are integrated into the HPC cluster.

At the same time, our computing cluster is being continuously expanded to ensure and strengthen Germany’s sovereignty as a centre for artificial intelligence in the future.

In this way, large models can also be trained and efficient proof of concepts and larger projects can be realised directly on site as part of our services.

fortytwo

42_Cluster
HPC Cluster with 632 A100-80GB-SMX GPUs

HPC Cluster with 79 Apollo 6500 server, each with

  • 2x AMD EPYC 7313 3.0GHz 16-core
  • 8x NVIDIA HGX A100 80GB GPU with NV-Link (SMX Card)   
  • 32x HPE 64GB Dual Rank x4 DDR4-3200 (= 2.048 GB)

HPC Graphcore Knot with

  • 2x AMD EPYC 7713 2.0GHz 64-core
  • 4x Graphcore bow-2000 Nodes
  • 16x HPE 32GB Dual Rank x4 DDR4-3200 (= 512 GB)

Parallel File Storage System with

  • 1.251 TB of usable capacity
  • 192 GB/s read, and 152,3 GB/s write

Maschine Learning Development Environment

Our cluster offers a unique interface for training and evaluating AI models. HPE’s Machine Learning Development Environment provides a standardised interface with WebGUI and command-line interface for easy integration of our cluster into your development processes. MLDE reduces the complexity of training, allows for infinite scaling of the experiment to up to 632 GPUs and simple collaboration between geographically distributed teams without major adaptations in the model code. For a first insight into our development environment, take a look at our OnBoarding video.

Documentation & knowledge base

Below you will find a collection of the most important information on the efficient use and independent troubleshooting of our computing infrastructure, as well as a link to our service portal for more detailed questions.

Access to the cluster

Who can apply to use our computing power?

The use of our services is open to all companies and institutions, unfortunately only private individuals cannot be offered use. It is possible to use our services as part of a proof of concept sprint as well as a co-operative small, medium or large project.

How can an application for the use of computing power be submitted?

The application for the use of computing power can be submitted under this link. Please select the application “Apply for HPC Cluster”

What are the requirements for a successful project application?

All that is required for a successful application for a small project is a fully developed project description with scientific added value and an appropriate project team. Medium and large projects, on the other hand, require a previous proof of concept and / or corresponding previous studies / publications.

Is there a limit to the utilisation of the computing power?

As part of the application for the use of computing power, please state the number of GPUs required and the expected project duration. A committee of 3 professors and 2 technical contacts will decide on this application. There is only a limit to the maximum number of GPUs that can be allocated in the form of hardware availability, but the maximum project duration is limited to 12 months.

Utilisation of the cluster

How can I use the cluster and train models there?

In our onboarding video, we explain the first steps for using our cluster in detail. You can find more in-depth information in our knowledge base and below in the “Machine Learning Development Environment” section.

Note: Once your application has been approved and your user has been created on our cluster, you will receive access to our knowledge database. Access is only possible for active cluster users.

Terms of use and fees

Information will follow…

Machine Learning Development Environment

Is there a collection of best practices?

A collection of tutorials, best practices and examples can be found in our knowledge base.

Note: Once your application has been approved and your user has been created on our cluster, you will receive access to our knowledge database. Access is only possible for active cluster users.

How can I start experiments and JupiterLabs on the cluster?

Our Machine Learning Development Environment from Determined.ai provides the basis and interface for cluster utilisation. This platform offers a wide range of functions and training options and enables almost infinite scaling of AI models.

Detailed documentation of the interfaces and functions can be found at https://docs.determined.ai/latest/.

What do I do if I have questions or something doesn’t work?

In the event of questions, problems and uncertainties, our service portal offers a centralised interface for answering your questions and support from our experts.

You can find our service portal at: https://hessian-ai.atlassian.net/servicedesk/customer/portal/3

Note: Once your application has been approved and your user has been created on our cluster, you will receive access to our service portal. Access is only possible for active cluster users.