Rethinking Data Infrastructure Conventions for Modern AI and HPC

HPC has long been optimized for one goal: run a single large application as fast as possible. Large-scale AI changes that. Training, fine-tuning, and inference each carry unique computational motifs and data access patterns, and blending these workloads with traditional simulation means questioning assumptions the HPC community has held for decades. Join us to explore what needs to change and why.

What you will walk away with:

• A clear understanding of how training, fine-tuning, and inference differ from traditional simulation in their use of data infrastructure

• Insight into which long-held HPC design conventions no longer hold when large-scale AI workloads enter the picture

• Why HPC's infrastructure design metrics fail AI workloads, and what questions to ask instead