Skip to content

GPU Access

Jupyter4NFDI now offers GPU-enabled resources for compute-intensive tasks such as machine learning, deep learning, and scientific computing. However, GPU resources are currently limited and require special access.

GPU Availability

Currently, GPU access is available on a limited number of GPU-enabled nodes. These resources are shared among all users and allocated on a first-come, first-served basis.

Available GPU Profiles

When GPU access is enabled for your account, you may select from the following GPU profiles:

  • JSC-Cloud: 1 GPU (Nvidia A100 80GB PCIe), 8GB RAM, 2 VCPU
  • deNBI-Cloud: 1 GPU (Nvidia Tesla V100 PCIE 16GB), 8GB RAM, 2 VCPU

On JSC-Cloud the we're using the Nvidia Multi Instance GPU feature. On deNBI-Cloud we're using a time-sliced approach, since MIG is not supported for the V100 GPU.

How to Request GPU Access

Due to the limited availability of GPU resources, access must be requested individually. To request GPU access:

  1. Contact us via email at jupyter4nfdi at lists.nfdi.de
  2. Include the following information in your request:
  3. Purpose of GPU usage (e.g., machine learning, deep learning, specific research project)
  4. Expected duration of GPU usage
  5. Any specific requirements or questions

Approval Process

After submitting your request, our team will review it and typically respond within 1-2 business days. Once approved, you'll receive an email invitation from Helmholtz ID to a virtual organization. Joining this virtual organization will grant you access to the GPU flavors.

Support

For questions about GPU access or to request access:

  • Email: jupyter4nfdi at lists.nfdi.de
  • Technical Support: ds-support at fz-juelich.de
  • Chat: NFDI Mattermost (Channel #jupyter4nfdi)

Future Plans

We are actively working on expanding GPU availability to all users without requiring individual access requests. This includes:

  • Adding more GPU nodes to our infrastructure
  • Implementing automated access control and resource management
  • Improving GPU resource scheduling and allocation

Our goal is to make GPU resources readily available for all users who need them for their research and projects.