BLACK FRIDAY EARLY ACCESS

25% off every PCTaken off the full price. Ends Cyber Monday.

ENDS IN

Shop early access
Skip to content
  • FAST AUSTRALIA-WIDE DELIVERY
  • USUALLY DELIVERED IN 1–3 WORKING DAYS
  • DELIVERY FROM MELBOURNE
  • 3-YEAR WARRANTY
  • CUSTOM BUILT TO ORDER
  • QUALITY TESTED PCS

GRAPHICS & AI · 5 MIN READ

What Is an NPU? AI Processors in PCs Explained

The small AI engine now built into many processors, what it actually does, how it compares with your graphics card, and whether it matters for you.

NPU explained: the neural processing unit inside modern PCs

An NPU is a small, specialised part of a processor designed for one job: running artificial intelligence tasks efficiently. It sits alongside the familiar CPU cores and graphics on many newer chips from Intel, AMD and Qualcomm, and it’s the reason you’ll now see “TOPS” figures on spec sheets next to clock speeds and core counts.

So what does it actually do, how is it different from your graphics card, and does it matter when you’re choosing a computer? Here’s the plain-English version.

NPU meaning: neural processing unit

NPU stands for neural processing unit. “Neural” refers to neural networks, the type of software behind most modern AI, from voice recognition to image generation. Running those networks involves enormous numbers of simple multiplications and additions. A CPU can do that maths, but slowly and with a lot of power. A neural processing unit is built to do almost nothing else, so it does it quickly while sipping power.

That’s the short NPU meaning: a dedicated, power-efficient engine for AI tasks, so the rest of the chip doesn’t have to do the heavy lifting.

CPU vs GPU vs NPU

CPU GPU NPU
Built for General tasks, one after another Graphics and huge parallel workloads AI tasks, efficiently
AI performance Low Very high Moderate
Power use for AI High High Very low
Best at Running Windows and apps Games, rendering, big AI models Always-on features, battery life

The key thing to understand is that a modern graphics card is far more powerful at AI than any built-in NPU. An RTX graphics card can run large AI models, image generators and video tools that an NPU simply can’t handle. What the NPU offers is efficiency: it can run lighter AI features in the background for hours without draining a laptop battery or spinning up fans.

Our guide to what a CPU does covers the main processor if you’d like a refresher on how they fit together.

What is TOPS?

TOPS means trillions of operations per second, and it’s the headline number for AI performance. Higher is faster. Microsoft’s Copilot+ PC programme, for example, requires a chip with at least 40 TOPS from its neural engine, along with 16GB of memory, to unlock certain Windows features.

Be careful comparing TOPS across brands and products, though. The figure can be measured in different ways, and some companies quote a combined number for the CPU, graphics and neural engine together. Real-world performance depends just as much on software support.

What does an NPU actually do today?

Here are the most common uses on Windows right now:

  • Video call effects. Background blur, eye contact correction and automatic framing in Windows Studio Effects run on the neural engine, so they don’t slow down the rest of your PC.
  • Noise reduction. Cleaning up microphone audio in calls.
  • Live captions and translation. Copilot+ PCs can caption and translate audio in real time.
  • Photo and creative tools. Some editing apps use it for features like subject selection and upscaling.
  • Smarter search and assistants. Windows features that understand what’s on screen or in your files.

More apps are adding support every year, but it’s still early. Many AI features in popular software use the graphics card or the cloud instead.

Which processors have one?

Most recent laptop chips include a neural engine, and many meet the 40 TOPS bar for Copilot+ features, including chips from Qualcomm’s Snapdragon X range, AMD’s Ryzen AI laptop processors and Intel’s newer Core Ultra laptop chips.

On desktops, it’s more mixed. Intel’s Core Ultra desktop processors include a smaller NPU processor block, good for features like video call effects but below the Copilot+ threshold. AMD’s Ryzen 9000 desktop chips don’t have one at all. Desktop users who want serious AI performance rely on their graphics card instead, which is far more capable anyway.

Wikipedia’s overview of the neural processing unit has more on how different companies build them.

Does your next PC need an NPU?

It depends on the kind of computer:

  • Laptops: yes, it’s worth having. Battery life during video calls and background AI features is noticeably better, and you’ll get access to Copilot+ features if the chip meets the bar.
  • Office and home desktops: nice to have. Intel’s Core Ultra desktop chips include one, which helps with video call effects without loading the main cores.
  • Gaming and creative desktops: not important. Your graphics card does AI tasks much faster, from DLSS upscaling in games to AI tools in editing apps.

If you run local AI models, like chatbots or image generators on your own machine, focus on the graphics card and its memory. That’s where the performance is, and it’s not close.

Privacy and on-device AI

One genuine advantage of running AI on your own hardware is privacy. When a feature runs locally, your audio, images and documents don’t need to be sent to a cloud service to be processed. That’s appealing for business users and anyone handling sensitive files. It’s also quicker for small tasks, with no waiting on an internet connection.

The trade-off is capability. The biggest AI models still live in data centres, and local hardware handles smaller, more focused jobs. Expect a mix of both for some time yet, with your PC handling the quick, private tasks and the cloud taking on the heavy ones.

What to look for on a spec sheet

  • TOPS figure: check whether it’s for the neural engine alone or the whole chip combined.
  • Memory: 16GB is the minimum for Copilot+ features, and 32GB gives more room for AI apps.
  • Graphics card: for creative AI tools and local models, the card’s own memory matters most. 12GB or more is a good target.
  • Windows version: the latest Windows 11 updates unlock the newest features.

The bigger picture

Neural engines are part of a broader shift, with chipmakers building AI acceleration into every layer of a PC. Over the next few years, expect more everyday software to quietly use whatever AI hardware your computer has, picking the most efficient option for each job. A machine with a modern processor and a capable graphics card is well placed for whatever comes next.

These desktops use Intel Core Ultra processors with a built-in neural engine, and they pair it with plenty of memory and fast storage. For a quiet workspace to put one in, our home office setup guide has ideas:

NPU questions

Is an NPU the same as a GPU?

No. Both can run AI, but a graphics card is far more powerful and uses more power. A neural engine is smaller and much more efficient for light, always-on tasks.

Can I add an NPU to my PC?

Not as a separate part for typical home PCs. It comes built into the processor. A graphics card is the practical way to add AI performance to a desktop.

Does an AI engine drain a laptop’s battery?

The opposite. Moving AI tasks off the main cores and graphics usually saves power, which is the whole point of having one.

Do games use the NPU?

Not really. Gaming AI features like DLSS run on the graphics card’s own AI cores.

What does NPU processor mean on a spec sheet?

It means the chip includes a neural processing unit alongside its regular cores, usually with a TOPS figure showing how fast it is.

// BUILT IN MELBOURNE

Want a PC that's ready for all of this?

Every build is put together to order, tested under load and delivered Australia-wide.

Shop PCs Ask us a question