Research Data for LLM-Based GPU Power, Performance, and Thermal Footprint Prediction Experiments

The research characterizes five deep-learning architectures across NVIDIA Tesla T4 and Quadro GV100 GPUs, evaluates large-language-model-based GPU power predictions, and investigates the effects of hardware power capping on power consumption, throughput, thermal behavior, and energy per completed training step.

The study also includes controlled T4 power-cap experiments and cross-hardware validation on NVIDIA L4 GPUs. The data are intended to support the reproducibility and verification of the experimental results reported in the associated research article.</p>