Workshops Archive

Wafer Scale Acceleration of Unstructured Finite Element Methods


Workshop: Twelfth International Workshop on Heterogeneous High-Performance Reconfigurable Computing (H2RC 2026)

Authors: Emma Urquhart (University of Cambridge)

Abstract: Finite element simulations on unstructured meshes require high performance and bandwidth to solve realistic problems. The Cerebras Wafer-Scale Engine (WSE) represents a paradigm shift in large-scale compute acceleration, integrating 900,000 processing elements (PEs) on a single device. While prior work has demonstrated strong performance for structured stencils on the WSE, FEM on unstructured meshes remains largely unexplored, owing to the dissonance between its irregular connectivity patterns and the WSE's structured dataflow model. We present an architectural mapping that reconciles the disparate mesh structures of the target architecture and application. For the heat equation, we demonstrate near-ideal weak scaling up to the wafer scale and sub-linear cost with increasing per-PE workload, showing the architecture's advantages for structured stencils extend to irregular connectivity. We establish unstructured FEM as a workload that wafer-scale architectures can serve, and identify per-PE capacity, rather than irregularity, as the constraint that governs how far the approach extends.


Back to Twelfth International Workshop on Heterogeneous High-Performance Reconfigurable Computing (H2RC 2026) Archive Listing

Back to Full Workshop Archive Listing