The Evolving Landscape of HPC for Weather, Water and Climate - a view from HPE

We’re now in a world where the largest computing systems are constructed to train AI models rather than to support scientific computing. Architectural specialization has shifted to support AI workloads rather than general-purpose or traditional scientific computing needs. Demand for the latest GPUs, as well as data center space to house them, has exploded and prices for both have risen well beyond what HPC have become accustomed to.

The CPU space is also increasingly complex, with very large numbers of SKUs for some of the leading processors and variation in core types (dense/classic, P-cores vs. E-cores, Arm, etc.). At the same time, performance increases, for both CPUs and GPUs, come at the expense of increased operating power, which puts additional burdens on budgets and facilities.

How can HPC centers that support weather, water and climate workloads navigate this dynamic computing landscape? What should be optimized, and how? This talk will discuss pitfalls of yesterday’s approach applied to today’s realities and discuss potential new approaches.