2026 Overview of HPC in Australia

Daniel Rodwell, Associate Director of Storage at Australia's National Computational Infrastructure (NCI), presents a joint overview of NCI in Canberra and the Pawsey Supercomputing Centre in Perth, whose flagship systems Gadi (2019) and Setonix (2021-22) are approaching end of design life. He explains the NCRIS funding scheme covering 26 national research capabilities and the National Digital Research Infrastructure (NDRI) strategy, which has delivered 55 million Australian dollars to NCI and 22 million to Pawsey for HPC refresh, a data federation project addressing persistent storage co-located with HPC, and a further 32 million for AI and machine learning compute, bringing NCI's current procurement to roughly 80 million. Stakeholder consultations reveal consistent themes: strong ongoing demand for CPU capability alongside rapidly growing GPU demand for both simulation and AI/ML, insistence that AI capacity be integrated within HPC centers rather than separate, and the need for performant co-located storage that avoids copying data into place. Responses varied on FP64 GPU support, CUDA and architecture choices, checkpoint restart, and maintenance sensitivity. A recent addition of 120 H200 GPUs to Gadi filled immediately. Rodwell also surveys institutional systems, including a new GB200 NVL72 platform at Monash University, the University of Queensland's Bunya system with its evergreen refresh philosophy, and UNSW's Katana, before closing with the need for a decadal plan to keep Australian research internationally competitive.