Skip to the content.

About

State of Practice in Deploying Supercomputers with NVIDIA Superchips (SPIN-NVSC) Workshop will be held during the 55th International Conference on Parallel Processing (ICPP’26), held in Singapore, September 28 - October 1, 2026. SPIN-NVSC is a full day workshop taking place on Monday, September 28.

HPC centres worldwide are navigating a significant architectural transition, as they seek to balance the demands of traditional simulation workloads with the rapid growth of AI and data-intensive applications. The emergence of unified CPU-GPU Superchip platforms — spanning the Grace-Grace, Grace Hopper, Grace Blackwell, and forthcoming Vera Rubin architectures — offers a coherent family of solutions targeting different points in this workload spectrum, from memory-bandwidth-bound simulation to large-scale AI inference and training. With early adopters now deploying these systems into production and Vera Rubin entering availability in the second half of 2026, the community is actively navigating opportunities and challenges spanning system design, software stack integration, workflow portability, application readiness, resiliency, resource management, and energy efficiency. This workshop, organised by the NVIDIA Supercomputing Users Group (NVSUG), brings together HPC practitioners, system architects, and computational scientists deploying or planning to deploy systems based on these architectures. Contributions from application developers and end users are strongly encouraged.

Topics

Key topics include, but are not limited to:

Workshop Schedule

The timing of the schedule will be announced when the conference organizers provide information about the coffee/meal breaks. The schedule will include:

Key Dates

Submissions

SPIN-NVSC is accepting short papers (6 pages) and long papers (10 pages). The page limit includes all references and appendices.

SPIN-NVSC workshop uses the ICPP Linklings for submissions.

To submit, please use this link and select the NVSUG submission type. For other submission rules (e.g., blind review rules), please refer to the general ICPP guidelines that the workshop will use.

Committee

Co-Chairs

Individual Affiliation
Sadaf R. Alam Bristol Centre for Supercomputing
Taisuke Boku Advanced HPC-AI R&D Support Center
Maciej Cytowski Pawsey Supercomputing Research Centre

Program Committee

Individual Affiliation
John Cazes TACC
Brandon Cook LBNL NERSC
Ryohei Kobayashi Institute of Science Tokyo
Adam Lavely LBNL NERSC
Colin McMurtrie CSCS/ETH Zurich
Jorge Luis Galvez Vallejo National Computational Infrastructure
Ugo Varetto Pawsey Supercomputing Research Centre
Wang Yi NSCC
Himeshi De Silva NSCC
Todd Evans NVIDIA
Simon See NVIDIA
Cerlane Leong CSCS

Tentative Agenda

Monday, September 28, 2026

Time Talk
09:00 – 09:10 Welcome
09:10 – 10:00 Keynote Session Chair: Taisuke Boku

Surrounding Challenges of GPU-Centric Supercomputing: IO, Energy Efficiency, and Operation, Toshihiro Hanawa, The University of Tokyo
10:00 – 10:30 NVIDIA Supercomputing User Group (NVSUG) Community Update Colin McMurtrie
10:30 – 11:00 Break
11:00 – 12:30 Full Papers Session Session Chair: Colin McMurtrie

Portable and Resilient by Design: A Cloud-Style Stack for HPC and AI across NVIDIA Grace and Grace Hopper Superchips, Sadaf R. Alam, Simon McIntosh-Smith

Scaling the Extended OpenDwarfs: Evaluating Cross-Vendor CUDA Performance Portability with SCALE, Beau Johnston, Chris Kitching, Matthew Ireland, Michael Søndergaard

Characterizing Communication Compression on NVIDIA BlueField-3 DPU for Distributed Training, Taiga Kobayashi, Tomohiro Ueno, Ryohei Kobayashi
12:30 – 13:30 Lunch
13:30 – 14:40 Panel Discussion GPU Coding in the AI Era, Moderator: Taisuke Boku, Panelists: Sadaf Alam, Tomohiro Ueno, Toshihiro Hanawa
14:40 – 15:00 Short Papers Session Session Chair: Ryohei Kobayashi

Beyond the HBM Wall: Coherent-Memory Oversubscription for AI Workloads on NVIDIA Grace Hopper, Mohamed Abdelnaby, Wu-chun Feng
15:00 – 15:30 Break
15:30 – 16:30 Short Papers Session Session Chair: Colin McMurtrie

Grace or Hopper? A Case Study of Wait-Free Concurrent Queues on the NVIDIA Grace Hopper GH200 Superchip, Pratheek Prakash Shetty, Atharva Gondhalekar, Wu-chun Feng

Adaptive GPU Sharing for Real-time LLM Serving with Best-effort Workloads, Shuxin Li, Toshihiro Hanawa, Yohei Miki

A Preliminary Study on Simultaneous Coscheduling for Discrete GPU vs. Fused GPU, Poorna Gunathilaka, Kirshanthan Sundararajah, Wu-chun Feng
16:30 – 16:45 Closing Remarks