# NPU

> A small AI engine inside phones and laptops that runs models fast on little battery.

- Level: Intermediate
- Topics: Hardware, AI
- URL: https://tidbit.legios.com.ar/npu/
- In Spanish: [NPU](https://tidbit.legios.com.ar/es/npu/)

An **NPU** (neural processing unit) is a part of a chip dedicated to running AI models. Modern phones and many recent laptops have one inside their main chip, next to the CPU and the GPU. It is what blurs the background in a video call, finds faces in your photo library, transcribes a voice note or runs a small language model without sending anything over the internet.

Like a TPU, it is specialized in the multiplications that neural networks are made of, using low-precision numbers. The difference is the goal. A TPU lives in a data center and trains or serves giant models; an NPU lives in your device and runs small models for inference using as little energy as possible. The same task on the CPU would be slower and drain the battery faster.

NPUs matter because of **on-device AI**: processing data on the device itself is quick, works offline and keeps private things like photos and voice off someone else's server. Manufacturers often advertise NPU speed in TOPS (trillions of operations per second), but that number says little on its own; what counts is whether the apps you use actually take advantage of it. Brand names vary too: Apple, for example, calls its NPU the Neural Engine.

## An analogy

A pocket translator versus phoning an interpreter. The interpreter far away can handle anything, but the little device in your pocket answers instantly, works without signal and keeps the conversation to yourself.

## Keep in mind

- A part of a phone or laptop chip made for running AI models.
- Runs small models locally, quickly and on little battery.
- Keeps data on the device, so some AI features work offline and more privately.
- TOPS figures in ads are a rough guide, not a measure of real usefulness.

## Go deeper

- [Neural processing unit on Wikipedia](https://en.wikipedia.org/wiki/Neural_processing_unit)
- [What is a neural processing unit? (IBM)](https://www.ibm.com/think/topics/neural-processing-unit)

## Related terms

- [Chip](https://tidbit.legios.com.ar/semiconductor-chip/): A sliver of silicon carrying billions of tiny switches that compute and remember.
- [Artificial intelligence](https://tidbit.legios.com.ar/artificial-intelligence/): Getting computers to do things that, until recently, only people could do.
- [CPU](https://tidbit.legios.com.ar/cpu/): The chip that runs your programs, one instruction after another, very fast.
- [GPU](https://tidbit.legios.com.ar/gpu/): The chip that does thousands of simple calculations at once.
- [LLM](https://tidbit.legios.com.ar/llm/): A neural network trained on huge amounts of text to predict the next word.
- [Internet](https://tidbit.legios.com.ar/internet/): A worldwide network of networks that lets any connected device reach any other.
- [TPU](https://tidbit.legios.com.ar/tpu/): Google's chip designed specifically for the math of neural networks.
- [Neural network](https://tidbit.legios.com.ar/neural-network/): A pile of tiny math units that learns patterns from examples.