Scientific research organisations are asking their HPC resources to do something new: run AI training and inference alongside traditional simulation, on the same data, often within the same power envelope.
The result is friction in places many teams didn’t expect.
Data sits in silos across cloud, edge and on-premise systems, so GPUs wait on information that exists but cannot be reached. Power and cooling limits put a cap on how far clusters can scale. And CPU- and GPU-centric architectures are being pushed hard by workloads they weren’t designed to mix.
In this panel, experts in data orchestration, grid-aware energy management and next-generation compute architecture discuss where hybrid AI-HPC environments are breaking down, what the practical fixes look like and where research organisations should invest first.
Expect a candid discussion of the trade-offs, plus a live Q&A, moderated by Scientific Computing World editor, Robert Roe.
