Abstract
Data centers and their compute nodes require accurate and flexible digital twins capable of modeling the complex interplay of workloads, environmental parameters, and physical metrics. Current machine learning approaches for HPC and its telemetry typically rely on a static subset of anonymous, fixed-position sensor variables tailored to single tasks. Consequently, these models become obsolete when target tasks change or sensor metrics vary. We propose SeT-Diff, the first foundational model for compute node telemetry and time-series. Unlike rigid architectures, our diffusion-based approach conditions the generative process on each sensor's semantic description, decoupling the system dynamics from the structure of the dataset.
Experiments on a real-world supercomputer dataset demonstrate a Mean Absolute Error (MAE) of 0.0470 on reconstruction tasks. SeT-Diff exhibits zero-shot permutation stability, maintaining accuracy with negligible degradation even when sensors are shuffled. A single pre-trained model effectively performs data imputation, forecasting, and virtual sensing—achieving a 0.033 MAE in thermal inference—making SeT-Diff an effective data-driven digital twin for HPC systems.
Blogger's Review: The introduction of SeT-Diff marks a significant advancement in the field of high-performance computing. Its diffusion-based approach not only enhances model flexibility and accuracy but also addresses the limitations of traditional methods when faced with dynamic tasks. In the future, this model's application will greatly enrich telemetry capabilities and intelligent decision support within HPC systems.