263 words
1 minute
DHinfra at ASHPC'26: talk and poster on federated GPU infrastructure and managed LLM inference

At ASHPC’26 (8–10 April 2026, TUtheSky, TU Vienna) we contributed two pieces on behalf of the DHinfra.at team at the Department of Digital Humanities, University of Graz:

  1. Talk (Technical Track): “Design and Operation of a Federated GPU Cluster for Digital Humanities within DHinfra.at” — Florian Atzenhofer-Baumgartner, David Fleischhacker, Max Resch, Lukas Waldhofer, Michael Otto. Slides (public version).
  2. Poster: “A Domain-Aware Controller for Managed LLM Inference on Shared HPC Infrastructure in Digital Humanities” — Michael Otto, Lukas Waldhofer, David Fleischhacker, Max Resch, Florian Atzenhofer-Baumgartner.

Booklet of Abstracts (DOI: 10.25365/phaidra.765, ISBN 978-3-200-10998-8).

Highlights#

From our DHinfra perspective, the most relevant presentations (roughly in order of relevance to our work):

  • Katrin Muck — “ASC Cluster Admin & Infrastructure Service Modernization”
  • Marko Ferme — “Bridging HPC and Cloud: OpenStack-Based Infrastructure for Efficient AI Computing”
  • Adam McCartney — “The MUSICA software stack”
  • Sebastian Sitkiewicz — “HPC Info: Enhanced SLURM Job Resource Monitoring”
  • Georg Heiler — “From Laptop to Supercomputer: Reproducible ML Pipelines with dagster-slurm and metaxy”metaxy
  • Hernan Picatto — “dagster-slurm: Connecting Data Orchestration to HPC Resources”dagster-slurm
  • Ümit Seren, Leon Schwarzäugl — “Zero-Touch HPC Nodes: NetBox, Tofu and Packer for a Self-Configuring SLURM Cluster”
  • Thomas Rattei — “LiSC software catalog: a software installation framework for Life Sciences”
  • Florian Goldenberg — “Moving stuff around: Rucio & FTS”
  • Gent Rexha — “Managed Multitenant ML Workflows on HPC”
  • Iulia-Georgiana Rinea — “Managed ML Inference on Shared HPC Infrastructure”

Heiler’s and Picatto’s talks in particular were technically solid, well-argued, and aligned with the push for a sovereign, barrier-lowering tech stack for supercomputing users.

It was an impressive range of HPC use cases overall, with a consistent emphasis on sustainability and accessibility. More of the technical track, please!

DHinfra at ASHPC'26: talk and poster on federated GPU infrastructure and managed LLM inference
https://atzenhofer.github.io/posts/2026-04-10-ashpc26/
Author
Florian Atzenhofer-Baumgartner
Published at
2026-04-10
License
CC BY-NC-SA 4.0