GPU4U Pilot Project

Student-focused GPU access within DHInfra.at

The GPU4U project, originating from the University of Klagenfurt (AAU), is being launched as a pilot within the DHInfra.at infrastructure initiative. DHInfra.at is developing a national Machine Learning infrastructure primarily for Digital Humanities, with CLARIAH partners and DH research projects receiving priority access. GPU4U serves as an exploratory pilot to understand how computational resources might be utilized by a broader student community, while the infrastructure’s core mission remains focused on Digital Humanities research.

Addressing the GPU Access Gap

As part of this pilot exploration, GPU4U will test limited access to computational resources for select student projects. The following three use cases will be evaluated as the infrastructure is being set up:

Use Case 1: Supporting Resource-Intensive Student Projects

One workshop per semester where students from various faculties can present project ideas and apply for limited resource allocation. Approximately five selected projects per semester may receive temporary access to the DHInfra cluster (featuring 12x H200 GPUs and multiple L40s) for tasks such as LLM fine-tuning, VR simulations, or other computational workloads.

Use Case 2: Limited LLM Inference Access for Educational Purposes

One GPU may be allocated to provide controlled access to LLM inference (via Ollama and various models) through an API for specific educational use cases. This would enable experimentation with prompting strategies, context size, temperature, and other parameters within structured learning environments. The inference could be integrated into select courses, such as a pilot “Introduction to Databases” course, to explore how such tools might support learning in non-technical disciplines.

Use Case 3: Experimental VR-Based Teaching Support

This use case explores the potential for VR in teaching through streaming solutions, particularly in Game Studies and Engineering contexts, which could reduce dependency on individual high-performance workstations. Implementation would depend on local infrastructure availability, such as VR headset access, and remains subject to further evaluation.

This exploratory pilot allows DHInfra.at to better understand potential use cases beyond its core Digital Humanities mission, while maintaining its primary commitment to CLARIAH partners and DH research projects. The insights gained will inform future decisions about resource allocation and access policies.