3rd TPC Workshop @ ISC: Bridging the AI and HPC Gap
- Friday, 26 June 2026. 14:00 – 18:00 (2:00 – 6:00 PM).
- Venue: ISC in Hamburg, Germany (June 22-26, 2026)
Abstract
This workshop examines how to address the growing misalignment between frontier-scale AI models and the realities of contemporary high-performance computing systems. It focuses on the requirements for running trillion-parameter training, large-scale synthetic data pipelines, and multilingual scientific workflows reliably on modern supercomputers. The discussion emphasizes practical integration: how HPC schedulers, storage systems, networking, and software stacks must evolve to support foundation-model development, how to coordinate shared synthetic-data and benchmarking infrastructure under initiatives such as the Trillion Parameter Consortium; and how emerging AI techniques are beginning to assist HPC itself by enabling automatic code generation, performance optimization, and realistic workload simulation. By treating AI and HPC as mutually reinforcing domains, the workshop aims to articulate a concrete path toward systems, tools, and operational practices capable of sustaining the next generation of large-scale scientific computing.
This workshop provides a platform for experts from academia, government, and industry to explore these challenges, share recent advances, and discuss emerging opportunities in AI-driven scientific discovery. The workshop aims to bridge the gap between AI research and scientific applications by bringing together computer scientists, domain experts, and HPC practitioners to exchange ideas and foster collaboration. By showcasing cutting-edge AI/ML advancements, identifying key challenges in deploying AI at scale, and facilitating discussions on the future of AI in science, this workshop will serve as a catalyst for accelerating AI adoption in scientific research.
Co-Organizers
- Javier Aula-Blasco (Barcelona Supercomputing Center)
- Murali Emani (Argonne National Laboratory)
- Eduardo Iraola (Barcelona Supercomputing Center)
- Gokcen Kestor (Barcelona Supercomputing Center)
Agenda
14:00 – 14:10: Welcome and introduction.
14:10 – 15:00: Invited keynote (30-35 mins) Rio Yokota (Institute of Science Tokyo, Japan) on “Pre-training LLMs at Scale: Distributed Training, Sparse Experts, and Synthetic Data” + Moderated audience discussion (15-20 mins).
15:00 – 16:00: Themed roundtable. Pedro Valero-Lara (Oak Ridge National Laboratory, United States) + Yanfei Guo (NVIDIA) + Shreeya Badhe (C-DAC, India). Moderated by Gokcen Kestor (Barcelona Supercomputing Center, Spain).
16:00 – 16:30: Break.
16:30 – 17:50: Themed session. See list of accepted submissions below.
17:50 – 18:00: Closing remarks and adjourn.
Accepted presentations
- B. Hsu, C. Siebenschuh (ANL, United States), O. Gökdemir, B. Parrello, N. Getty, T. S. Brettin, R. L. Stevens, I. T. Foster, N. Chia & A. Ramanathan on “Generating Synthetic Biological Reasoning Problems at Scale on Heterogeneous HPC Systems”
- A. Navilarekal Rajgopal (LRZ, Germany) & N. Solmsdorf (Intel) on “A Scalable Recipe on SuperMUC-NG Phase 2: Efficient Large-Scale Training of Language Models”
- A. Oliveira-Filho, W. Silva-de-Souza, C. A. Valderrama Sakuyama & S. Xavier-de-Souza (UFRN, Brazil) on “Phoeni6: A Systematic Framework for Transparent and Reproducible Energy Profiling of Frontier-Scale AI”
- C. Siebenschuh (ANL, United States), B. Hsu, K. Hippie, A. Ramanathan, I. T. Foster & R. L. Stevens on “Scaling Structured Tensor Algebra for Time Series Foundation Model Training on Aurora”
- Y. K. Singh (C-DAC, India) on “Bridging Computation and Compassion: How C-FLOOD Transforms Disaster Response in India through HPC-AI Convergence”
Call For Submissions [Submissions Closed]
We welcome short one-pagers describing relevant work, ideas, systems, or experiences related to the topics of the workshop. Submissions will be reviewed by the workshop organizers. Accepted submissions will be invited for 15-minute presentations during the workshop.
Please use this form to submit your one-pager.
Submission deadline: May 8, 2026 AOE
Notification of acceptance: May 15, 2026
Workshop Topics
- Training Frontier-Scale AI Models on HPC Systems
- Challenges of doing pretraining, fine-tuning and inference of AI models on traditional HPC schedulers, storage hierarchies and network topologies.
- Software and runtime adaptations to support long-running, adaptive and communication-intensive foundation model workloads.
- Experiences and lessons learned from integrating model training and inference into supercomputing facilities.
- Synthetic Data Pipelines as HPC Workloads
- Turning large-scale, synthetic-data generation, curation and validation into reproducible HPC workflows.
- Strategies for multilingual and domain-diverse synthetic data production.
- Approaches to validate synthetic data quality at HPC-scale.
- Practical HPC Integration
- Execution models that better support AI workload, including job scheduling, resource allocation and data-movement.
- Efforts to align model architectures, training and inference strategies, and system characteristics.
- Case studies where foundation model development and inference required redesigns or policy changes at HPC facilities.
- AI for HPC Development
- Using AI models to assist in writing, refactoring and optimizing parallel and distributed code.
- Generating synthetic workloads, datasets, or performance traces to support system tuning, benchmarking and co-design.
- AI-driven orchestration, scheduling, and automation of complex HPC workflows.
- Performance evaluation and benchmarking of emerging AI accelerators (e.g., GPUs, TPUs, FPGAs, neuromorphic computing) for scientific and engineering workloads.
- Techniques for addressing scalability, optimization, and efficiency issues when deploying AI/ML on extreme-scale HPC platforms.
- Transparency, Reproducibility and Responsible Practice
- Methods to ensure traceability and reproducibility of frontier-scale AI runs and synthetic-data pipelines in HPC environments.
- Responsible resource usage, energy considerations and transparent reporting of computing consumption.
- Governance models for shared access to high-end compute infrastructure in Europe.