TPC Workshop Tackles Trillion-Parameter AI-HPC Gap at 3rd Hamburg Session
Updated
Updated · HPCwire · Aug 24
TPC Workshop Tackles Trillion-Parameter AI-HPC Gap at 3rd Hamburg Session
2 articles · Updated · HPCwire · Aug 24
Summary
June 26’s 3rd Trillion Parameter Consortium workshop in Hamburg centered on a widening mismatch between frontier AI models and current high-performance computing systems.
Rio Yokota’s keynote examined large-scale LLM pre-training, including distributed training, sparse expert models and synthetic data, framing the technical pressures behind that gap.
A roundtable asked whether AI Factories are HPC’s successor or an eviction notice, with speakers from Oak Ridge, NVIDIA and C-DAC offering competing views on infrastructure’s next role.
Five later presentations extended the debate to synthetic biological reasoning, scalable LLM training, energy profiling, code generation and reproducibility, before a closing discussion on responsible compute use and shared European governance.
EuroTPC, coordinated by the Barcelona Supercomputing Center, used the ISC High Performance conference to argue for coordinated European action as AI development outpaces existing HPC design.