Skip to content
Pioneer Centre for AI
Gallery

Book a consultation

Open in new tab

Book a consultation

Open in new tab
Event · D3A 4.0 2026

Supercomputing: Challenges & Agents

Thursday 8 October 2026, 11:00–12:30 · Salon 26+27, Hotel Nyborg Strand · D3A 4.0 2026 (session page).

Shared HPC is only worth it if researchers can actually get their work done on it. So we're bringing HPC providers and researchers into the same room to talk about what gets in the way, and to look at where agentic GenAI might help.

About this session

Format

  • 90 minutes, interactive
  • About a third talks
  • The rest is group work and discussion

Level

  • Intermediate
  • HPC experience helpful
  • No agent expertise required

Audience

  • AI/ML researchers
  • HPC operators
  • Domain scientists on shared compute

Programme

We start with three short talks, and then it's over to you. In groups, you write down the challenges you run into on the Discussion form, then read and rate what the other groups came up with on the Review form. We finish together with the issues you rated highest, and your questions.

  • Framing and plenary
  • Talk
  • Interactive
  1. 11:00 5 min

    Welcome and framing

    Benjamin Starostka Jakobsen

  2. 11:05 10 min

    Agentic HPC: GenAI and sustainability

    Claudio Pica · SDU eScience Center

  3. 11:15 10 min

    Operating HPC: Hosting research at scale

    Ulrik N. de Lichtenberg · DCAI / Gefion

  4. 11:25 10 min

    Shared HPC: Effective sharing of GPUs

    Pınar Tözün · IT University of Copenhagen

  5. 11:35 20 min

    Group discussion: challenges with shared compute

    All participants · Discussion form

  6. 11:55 20 min

    Group review: rate other groups' issues

    All participants · Review form

  7. 12:15 15 min

    Plenary: top-rated issues and audience Q&A

    Nicki Skafte Detlefsen · Impact board

  8. 12:30

    End of session

Outcomes

Questions we hope to answer together.

  1. Which frictions are blocking researchers from getting work done on shared HPC?
  2. What should HPC providers pick up next, and what useful practices can researchers share with each other in the meantime?
  3. As agents take over cluster work, will researchers still need cluster knowledge, and who debugs it when the agent gets it wrong?

Everything the groups write down ends up on the board below, and your ratings sort it into low, medium, and high impact. In the plenary we go through the high-impact issues together. The board fills up during the session, so it's worth keeping an eye on.

Open in a new tab

Board not loading? Some browsers block embedded content, but the new tab link above always works.

Organizers

Collaborations

Pioneer Centre for Artificial Intelligence Pan-Danish research centre advancing AI, machine learning, and computer vision. Technical University of Denmark DTU Computing Center provides centralised HPC services, software stacks, storage, and training to researchers across the university. Danish Centre for AI Innovation Home to Gefion, Denmark's NVIDIA DGX SuperPOD (191 DGX H100 nodes, 1,528 H100 GPUs) for large-scale AI projects.