Source · Tier A · Peer-reviewed

Single-Node Power Demand During AI Training: Measurements on an 8-GPU NVIDIA H100 System

I. Latif, A. C. Newkirk, M. R. Carbone, A. Munir, Y. Lin, J. Koomey, X. Yu, Z. Dong. 2025. IEEE Access, vol. 13, pp. 61740–61747.

Linkhttps://doi.org/10.1109/ACCESS.2025.3554728
DOI10.1109/ACCESS.2025.3554728
arXiv2412.08602
VersionPublished in IEEE Access under a new title. The arXiv preprint (v1 11 December 2024, v2 20 December 2024) is titled "Empirical Measurements of AI Training Power Demand on a GPU-Accelerated Node". The published abstract, checked 2026-09-24, states the 8.4 kW peak and the 18% gap to the rated 10.2 kW that citing records use.
Accessed2026-09-23
Imported fromhodgkins-ai-verification-papers@c71e59ff0e8e
NoteListed under "Power telemetry and side-channel attacks" in the Hodgkins bibliography (CC BY 4.0).

Cited by