What Is an NPU, and Does Your Laptop Need One?
What a neural processing unit does, how it differs from a CPU and GPU, what TOPS numbers mean, and why Copilot+ PCs require an NPU rated above 40 TOPS.
Short answer
An NPU (neural processing unit) is a processor block built for the matrix math inside AI models, so it runs tasks like background blur, live captions and on-device image tools using far less power than a CPU or GPU. You need one mainly for Copilot+ PC features, which Microsoft ties to an NPU capable of more than 40 TOPS.
Key takeaways
- An NPU handles AI inference efficiently; the CPU and GPU still do most of the general and graphics work.
- TOPS (trillions of operations per second) is a peak figure, usually measured at low precision such as INT8, so it is only a rough comparison.
- Microsoft's Copilot+ PC features require an NPU that can run at 40+ TOPS; older NPUs below that line miss those features.
Laptop spec sheets now list a third processor next to the CPU and GPU: the NPU. Chip makers quote it in TOPS, and Microsoft uses it to decide which PCs get its Copilot+ features. This post explains what the NPU actually does, how to read a TOPS figure, and when it should affect what you buy.
What does an NPU actually do?
An NPU is a dedicated block of silicon that runs the multiply-and-add math inside neural networks, doing it in parallel and at low power.
Most AI models, from speech recognition to image upscalers, boil down to huge numbers of matrix multiplications. A CPU can do this math, but it is designed for flexible, branching, general-purpose code. A GPU is much better at parallel math, but it draws a lot of power doing it. An NPU is narrower than both: it is built almost entirely around the operations neural networks need, often at reduced precision like 8-bit integers (INT8), which cuts the energy per operation.
The practical result is that sustained AI tasks, such as blurring your background on a video call for an hour, can run on the NPU while the CPU and GPU stay mostly idle. Microsoft's own support documentation describes the NPU as using less power and being more efficient at AI tasks than a CPU or GPU, freeing them for other work.
The NPU does not replace the other two. Training large models still happens on data center GPUs, and plenty of local AI software, including many tools for running AI models locally, uses the GPU because it has more memory bandwidth and wider software support.
How is an NPU different from a CPU and a GPU?
The CPU is the generalist, the GPU is the parallel workhorse for graphics and heavy compute, and the NPU is a low-power specialist for AI inference.
| Processor | Built for | Typical AI role | Power use for AI work |
|---|---|---|---|
| CPU | General code, operating system, app logic | Small models, preprocessing, fallback | Moderate to high |
| GPU | Graphics, large parallel workloads | Larger local models, image generation, training | High |
| NPU | Neural network inference at low precision | Always-on effects, captions, on-device assistants | Low |
On modern laptop chips, all three sit on the same package and share system memory. Windows can route a task to whichever processor fits it, provided the software was written to support that.
What does TOPS mean?
TOPS stands for trillions of operations per second, a theoretical peak rate for the NPU's math units.
A chip rated at 45 TOPS can, in ideal conditions, perform 45 trillion simple operations every second. Keep three caveats in mind when comparing numbers:
- Precision matters. Vendors usually quote TOPS at INT8. The same chip will post a lower number at higher precision such as FP16.
- Peak is not sustained. Thermal limits, memory bandwidth and driver quality decide how close real workloads get to the rated figure.
- Some marketing combines CPU, GPU and NPU TOPS into one "platform" figure. For Copilot+ eligibility, the NPU number is what counts.
TOPS is useful for sorting chips into rough tiers. It is not a benchmark of how fast a specific app will run.
What NPU do Copilot+ PCs require?
Microsoft states that many Windows AI features on Copilot+ PCs require an NPU capable of more than 40 TOPS.
According to Microsoft's Copilot+ PC developer guide, Copilot+ PCs are Windows 11 machines built around a high-performance NPU that can perform more than 40 trillion operations per second. Features designed for those NPUs ship with Windows 11 and run locally instead of in the cloud.
The first wave used Qualcomm's Arm-based Snapdragon X series. Microsoft's guide also lists Copilot+ PCs built on AMD Ryzen AI 300 series and Intel Core Ultra 200V series processors. Earlier laptops with first-generation NPUs, such as some Intel Core Ultra Series 1 models, have an NPU but fall below the 40 TOPS line, so they do not get the Copilot+ feature set. If the Arm part of that list makes you hesitate, our guide to Arm vs x86 laptops covers app compatibility.
Which devices have NPUs today?
As of November 2026, NPUs are standard in current Apple, Qualcomm, AMD and Intel laptop chips and in most flagship phone processors.
Apple calls its block the Neural Engine and has included it in iPhone chips for years and in every Apple silicon Mac. Qualcomm's Snapdragon phone and PC chips use its Hexagon NPU. AMD's Ryzen AI processors and Intel's Core Ultra lines include NPUs, and Google's Tensor chips in Pixel phones have their own AI accelerator. Names and capabilities differ, but the purpose is the same: run on-device AI without draining the battery.
To check your own machine, Windows 11 shows an NPU graph in Task Manager's Performance tab when a supported NPU and driver are present. Our guide on how to check your PC specs walks through the other places to look.
Should an NPU affect what you buy?
Buy for the NPU only if you want specific on-device AI features; otherwise CPU, RAM, battery and display still matter more.
An NPU is worth prioritizing if you want Copilot+ features in Windows, if you spend many hours on video calls with effects turned on, or if you use apps that explicitly list NPU acceleration. It matters less if your AI use is mostly cloud chatbots, which run on remote servers, or if you run larger local models, where GPU memory and system RAM are the limiting factors.
If you are choosing between two otherwise similar laptops, check three things: whether the NPU meets the 40+ TOPS mark, whether the apps you use support it on that platform, and how much RAM comes with the machine, since AI features also need memory to run.
Frequently asked questions
Can I see whether my PC has an NPU?
On Windows 11, open Task Manager and go to the Performance tab. Devices with a supported NPU and driver show an NPU graph next to CPU, Memory, Disk and GPU.
Do apps use the NPU automatically?
No. Software has to be written to target the NPU, usually through a framework such as Windows ML and ONNX Runtime on Windows. Apps that are not built for it keep running on the CPU or GPU.
Is a higher TOPS number always better?
Not always. TOPS is a theoretical peak at a given precision, and real speed also depends on memory bandwidth, drivers and how well a model is optimized for that specific NPU.
Do I need an NPU to use chatbots like ChatGPT?
No. Cloud chatbots run on the provider's servers, so any device with a browser works. An NPU matters for AI features that run locally on your machine.
Sources
- Copilot+ PCs developer guide (Microsoft Learn)
- All about neural processing units (NPUs) (Microsoft Support)
- Copilot+ PCs (Microsoft)
Research-based: Technorat does not claim hands-on testing. See how we research. Spotted an error? contact@technorat.com