Birds of a Feather (BoF) - SC25

Converged HPC-AI Platforms: Navigating the Challenges of Heterogeneous Systems

View this BoF in the official SC25 program

Session Leaders:
Clayton Hughes (Sandia National Laboratories),
Edgar A. Leon (Lawrence Livermore National Laboratory),
Patrick Carribault (CEA),
Julien Loiseau (Los Alamos National Laboratory)

Overview

Recent national laboratories and supercomputing centers are deploying heterogeneous systems that integrate multi-core CPUs with GPUs, AI accelerators, FPGAs, IPUs, and DPUs. These converged HPC-AI platforms promise unified infrastructures for simulation and AI workloads but introduce challenges in programming models, scalability, portability, and optimization.

This Birds of a Feather session explores co-design strategies to address these complexities, focusing on scalable frameworks, unified programming ecosystems, portable workflows, and tight hardware–software integration. Attendees will share experiences, discuss emerging architectures, and identify tools that enable effective convergence of HPC and AI.

Speakers

TBD — speaker information will be provided closer to the session date.

Key Themes

  1. Seamless Scalability: Achieving efficient scaling across diverse architectures.
  2. Unified Programming Ecosystems: Bridging HPC and AI with shared frameworks and models.
  3. Portability and Optimization: Maintaining performance across distinct architectures.
  4. Forward-Looking Co-Design: Preparing applications for post-exascale and AI-driven systems.

Session Format

The session will combine panel discussions with extensive audience interaction (approximately 75%), using the Slido app for live questions and polls. Topics include software–hardware co-design, system composition, and porting strategies. Short expert presentations will introduce each theme, followed by open discussion and audience input.

Session details:

  • Format: Panel of experts (vendor-neutral)
  • Duration: 1 hour (Thursday, 12:15 pm–1:15 pm)
  • Expected attendance: 90 participants
  • Keywords: Emerging Computing Technologies, Co-Design, Heterogeneous System Architectures

LA-UR-25-30174