DHinfra.at Public-Launch-Event
GPU Cluster and Large Language Models for the Humanities and Cultural Heritage in Austria
When: Monday, 12. October 2026, 10:00 - 11:30 CET
Where: online via Zoom - https://uni-graz.zoom.us/j/63796915567
Please join with your full name so that we can admit you from the waiting room. No prior knowledge is required, and there will be time for your questions at the end.
Language: The presentation is in German. Questions in English are welcome.
DHinfra.at (Digital Humanities Infrastructure Austria) is an initiative by nine Austrian universities and research institutions. From 2023 to 2026, we have collaboratively developed an infrastructure for the humanities and cultural heritage sciences: mobile multispectral camera systems, a scanning robot, databases and virtual servers for research projects, open software tools, and a distributed GPU cluster for machine learning and AI.
Starting October 1, project proposals will be accepted from anyone conducting research and teaching in the humanities, cultural heritage, and arts in Austria. During the launch event, we will first present the entire infrastructure, with a particular focus on the GPU cluster.
Further information about the GPU cluster can be found on the DHInfra website:
GPU-Cluster (DHinfra)What Makes the Cluster Unique
The cluster runs large open language and vision models with several hundred billion parameters. The selection of models changes over time: currently, Qwen 3.5 with 397 billion parameters is in operation. Future models will be chosen by user vote, with options including models from DeepSeek, Kimi, GLM, and Mistral. Additionally, smaller, faster models and embedding models are available. Researchers access these models through an interface compatible with common AI tools. Texts and images remain within the academic network and are not shared with any commercial providers.
Models of this scale are rarely available at universities. Those who wish to use them typically require expensive hardware or must send their data to commercial services. For archives, manuscripts, and unpublished sources, both options are often impractical. Until now, Austria lacked a GPU and AI infrastructure specifically designed for the humanities and cultural heritage sciences.
The cluster also provides computing time on NVIDIA GPUs (H200, RTX PRO 6000, L40S), JupyterHub for interactive work, as well as SSH and SLURM for larger computational tasks. Access is managed via eduID, using the digital account of the user’s home institution. Teaching activities and theses are also welcome: instructors describe their project in the application and invite their students collectively.
Although the cluster will officially begin regular operations on October 1, it will first enter an introductory phase lasting until December 31, 2026. During this period, selected projects can use the cluster free of charge. Plans for its operation beyond 2027 will be announced in due time by CLARIAH-AT and the DHinfra.at consortium.
What Is Already Running
Text recognition on historical newspapers, information extraction from medieval account books using language models, model training for the recognition of Latin inscriptions, and much more. During the launch event, we will showcase some of these examples.
Further detailed information about DHinfra.at can be found on the website:
DHinfra.at