GPU
Specialized hardware accelerator for parallel computations, widely used to train and run large-scale AI models efficiently.
Definition
Graphics Processing Units (GPUs) excel at matrix and tensor operations fundamental to deep learning. They dramatically reduce training and inference times but consume significant power and require specialized provisioning. Governance around GPU usage includes cost monitoring, job-scheduling policies, fair-share quotas, and retiring obsolete GPU generations to optimize performance-per-watt.
Real-World Example
A research lab provisions a dedicated GPU cluster for AI experiments. They enforce time quotas per user, automatically decommission models running beyond limits, and track GPU utilization metrics to justify upgrades to more energy-efficient GPU architectures.
.svg)
