This is an old revision of the document!
Table of Contents
About Us
The AI Service Centre Berlin Brandenburg is a project at the Hasso Plattner Institute. Our goal is to foster the use of artificial intelligence in business and society.
We support German start-ups, companies, public institutions, students, and researchers in developing and using AI applications. To this end, we provide the necessary resources, training, and advice at no cost.
Overview of the Infrastructure
Performance Characteristics
- Total performance: 5.4 PFLOPS (FP32)
- Main memory: 23.6 TiB
- Number and type of GPUs: 114 (64 H100, 40 A30, 8 L40S, 2 L40)
- Available storage: 1.5 PB
- Number of nodes + interconnect: 15 nodes, Infiniband NDR 400Gbps
System Overview
- H100 Nodes
- 8 nodes, each with 8 NVIDIA H100 (64 GPU)
- A30 Nodes
- 5 nodes, each with 8 NVIDIA A30 (40 GPU)
- L40S (GPU Server)
- 8 NVIDIA L40S (8 GPU)
- AMD Epyc CPU
- L40 (ARM server)
- 2 NVIDIA L40 (2 GPU)
- Ampere Altra Max 128-30 CPU, 2.8GHz
- Edge simulation
- NVIDIA Jetson AGX ARMv8 CPU
- Storage
- 1.5 PB VAST all-flash memory system
Using Our Computing Resources
Our infrastructure enables users to run AI workloads efficiently, whether for research or prototyping. We offer a heterogeneous infrastructure to support a variety of machine learning and deep learning tasks.
1. Make a Proposal
To initiate a pilot project, submit a proposal outlining your project goals, required resources, and expected outcomes. The proposal will be reviewed by the Steering Committee of the HPI AI Service Centre.
2. Account Setup
If your proposal is accepted, you will need to complete the following steps to set up your account:
- Accept the Terms of Use: English Terms of Use German Terms of Use
- Invite additional team members as needed.
The most important points from the terms of use in brief (disclaimer: this list does not replace reading the terms of use):
- Use of the resources is free of charge and we also cover the electricity costs via the AI Service Centre
- By using the computing resources, you agree to publish results and new created artefacts (e.g. code, datasets, models, learnings). To do so, you must submit a project report.
- You also commit to include the following funding notice in all further publications:
- DE: Das diesem Bericht zugrunde liegende Vorhaben wurde mit Mitteln des Bundesministeriums für Bildung und Forschung unter dem Förderkennzeichen »KI-Servicezentrum Berlin-Brandenburg« 01IS22092 gefördert. Die Verantwortung für den Inhalt dieser Veröffentlichung liegt bei der Autorin/beim Autor.
- EN: The project on which this report is based was funded by the Federal Ministry of Education and Research under the funding code “KI-Servicezentrum Berlin-Brandenburg” 01IS22092. Responsibility for the content of this publication remains with the author.
- Your intellectual property remains with you, but the HPI is granted a non-exclusive, transferable right of use for the purposes of research and education.
- You may only perform computations that you have mentioned in the project application.
3. Use the IT Landscape
Once your account is set up, you can start using the available hardware and software resources. Refer to:
- Configure an OpenVPN connection to access the infrastructure.
- Submit SLURM jobs to train AI models.
4. Document Results and Submit a Report
Throughout the project, document your findings and results. At the end of the pilot, submit a report using the provided template.
Report Templates:
5. Request an Extension (If Needed)
If more time or resources are required, submit an extension request for review.
For further assistance, contact the HPI AI Service Centre team.
Contact Us
If you have any questions or need assistance, consolidate our FAQ or feel free to reach out:
- For technical issues: Sci.Helpdesk(at)hpi.de
- For organizational issues: ki-servicezentrum(at)hpi.de
We look forward to supporting your AI projects!