The Austrian-Slovenian HPC Meeting (ASHPC) is an annual conference bringing together users and providers of high-performance computing. It serves as a growing regional platform for exchange between scientists, engineers, infrastructure providers, and HPC users across Europe.
This year’s edition, held in Vienna from 7 to 10 April, welcomed 160 participants from 10 countries. A central theme throughout the programme was the growing convergence of high-performance computing and artificial intelligence.
Across sessions, discussions ranged from energy-efficient HPC–AI deployments and large-scale model inference to reproducible machine learning workflows and emerging hybrid approaches involving quantum computing. The diversity of scientific applications, spanning astrophysics, climate science, life sciences, engineering, and digital humanities, highlighted the essential role of HPC in enabling complex, data-intensive research
Within this context, AI Factory Austria AI:AT contributed three presentations addressing key challenges at the intersection of AI and HPC infrastructure, workflows, and skills development.
Managed AI workloads on shared HPC systems
A central topic was the integration of machine learning workloads into traditional HPC environments. In her presentation on managed ML inference, Iulia-Georgiana Rinea (ML engineer) outlined how AI:AT is developing a platform to support real-time, multi-tenant inference services on shared GPU infrastructure.
Unlike conventional batch workloads, inference systems are continuous, latency-sensitive, and accessed via APIs. Addressing these requirements calls for new architectural approaches. The AI:AT platform combines Kubernetes-based orchestration with tools such as KServe and vLLM to manage model lifecycles and deliver efficient inference at scale. Particular emphasis is placed on fair resource allocation, including token-based metering, as well as comprehensive observability across the inference pipeline.
Particular emphasis is placed on fair resource allocation, including token-based metering, as well as comprehensive observability across the inference pipeline. The presentation also highlighted practical challenges, including GPU scheduling across heterogeneous systems and managing model cold-start times in a shared environment.
Simplifying ML workflows for researchers
Gent Rexha (Senior ML Engineer) presented the AI:AT software platform for managed, multi-tenant machine learning workflows on HPC systems. His work addresses a well-known gap: while HPC clusters offer substantial computational power, they often lack the user-friendly tooling expected in modern machine learning practice.
The platform introduces a layered architecture that abstracts the complexity of batch schedulers such as SLURM, integrates Kubernetes-based pipeline orchestration, and ensures tenant isolation across shared infrastructure. Researchers can define and execute workflows through a Python SDK, without requiring detailed knowledge of the underlying HPC systems
By combining reproducibility, scalability, and ease of use, the platform lowers the barrier for research groups adopting advanced AI methods while preserving the performance advantages of HPC.
Building skills across the AI ecosystem
Beyond infrastructure and tooling, skills development remains a key component of Europe’s AI strategy. In this context, Michael Iro (Coordinator of the AI:AT Learning Centre) presented the AI:AT Learning Center, which focuses on training, education, and knowledge transfer
As part of a broader European network of AI Factories, AI:AT contributes to strengthening the AI ecosystem by connecting compute resources, data, and expertise. The Learning Center’s activities include structured training programmes, self-learning platforms, curriculum development, and exchange opportunities. These initiatives support a wide range of stakeholders, from researchers and students to industry and public sector organisations.
Collaboration at the European level
ASHPC26 underscored the importance of European collaboration. Contributions from participants across the continent, alongside presentations from multiple EU initiatives, highlighted ongoing efforts to build a connected and interoperable AI and HPC landscape.
With ASHPC27 scheduled to take place from 6 to 9 April 2027 in Rimske Terme, the dialogue between HPC and AI communities in Europe continues to evolve.
Further information:
