Methodology for Fine-Grain GPU Power Visibility and Insights

Ubiquity of AI coupled with its steep power demands make optimizing GPU power a priority as large GPU-based clusters are often employed to train and serve AI models. An important first step in optimizing GPU power consumption is high-fidelity and fine-grain power measurement of key AI computations o...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Singhania, Varsha, Aga, Shaizeen, Ibrahim, Mohamed Assem
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Ubiquity of AI coupled with its steep power demands make optimizing GPU power a priority as large GPU-based clusters are often employed to train and serve AI models. An important first step in optimizing GPU power consumption is high-fidelity and fine-grain power measurement of key AI computations on GPUs. To this end, we observe that as GPUs get more powerful, the resulting sub-millisecond to millisecond executions make fine-grain power analysis challenging. In this work, we first carefully identify the challenges in obtaining fine-grain GPU power profiles. To address these challenges, we devise FinGraV methodology where we employ execution time binning, careful CPU-GPU time synchronization, and power profile differentiation to collect fine-grain GPU power profiles across prominent AI computations and across spectrum of scenarios. Using FinGraV power profiles, we make several observations pertaining to GPU power variation over executions and over time, GPU sub-component power consumptions across different scenarios, and power behavior over interleaved executions of multiple computations. Equipped with these observations, we conclude with several recommendations to optimize the power for these ubiquitous accelerators.
DOI:10.48550/arxiv.2412.12426