PRISM: Dynamic Primitive-Based Forecasting for Large-Scale GPU Cluster Workloads
This addresses efficient scheduling and resource management for GPU-powered AI platforms, representing a domain-specific incremental improvement.
The paper tackles the problem of forecasting volatile, heterogeneous GPU workloads in AI infrastructure by proposing PRISM, a primitive-based compositional forecasting framework that achieves state-of-the-art results on large-scale production traces and significantly reduces burst-phase errors.
Accurately forecasting GPU workloads is essential for AI infrastructure, enabling efficient scheduling, resource allocation, and power management. Modern workloads are highly volatile, multiple periodicity, and heterogeneous, making them challenging for traditional predictors. We propose PRISM, a primitive-based compositional forecasting framework combining dictionary-driven temporal decomposition with adaptive spectral refinement. This dual representation extracts stable, interpretable workload signatures across diverse GPU jobs. Evaluated on large-scale production traces, PRISM achieves state-of-the-art results. It significantly reduces burst-phase errors, providing a robust, architecture-aware foundation for dynamic resource management in GPU-powered AI platforms.