GPU
Also known as: graphics processing unit
A GPU is a chip designed to do enormous amounts of simple math in parallel. That same parallelism is exactly what neural networks need — which is why GPUs became the default AI hardware.
- Training: thousands of GPUs working together, for weeks.
- Inference: one or a few GPUs serving requests.
You’ll hear “GPU” used interchangeably with “AI accelerator” in casual talk.
Local relevance: the GPU in your laptop/Mac is what runs local models. More VRAM = bigger models you can run. That's the whole local story in one sentence.