COMPEL Glossary / gpu-graphics-processing-unit
GPU (Graphics Processing Unit)
A GPU is a specialized processor originally designed for rendering graphics in video games, now widely repurposed for AI workloads.
What this means in practice
GPUs contain thousands of small cores that can perform many calculations simultaneously (parallel processing), making them ideally suited to the matrix operations that neural networks require. NVIDIA dominates the enterprise GPU market with its A100, H100, and subsequent generations. For transformation leaders, GPUs have significant budget implications: a single high-end GPU can cost over $30,000, cloud GPU instances range from $1-30+ per hour, and training a large model can consume thousands of GPU-hours. GPU costs represent a growing portion of AI budgets that must be factored into every business case and managed through AI FinOps practices.
Related Terms
Other glossary terms mentioned in this entry's definition and context.