GPU

Last updated Jul 17, 2026

Graphics Processing Unit — hardware that **parallelizes matrix math**, making neural network inference and training much faster than CPU alone.

#When to use

Running local LLMs, training models, or high-throughput serving.

Tiny batch jobs where CPU + API is cheaper and simpler.

A 7B parameter local model feels usable on a GPU; on CPU alone it may be slow for chat.