Trace Id is missing
Skip to main content
Surface
A detailed motherboard bristles with a central processor, capacitors, resistors, and circuits

May 28, 2024

What are NPUs?

Artificial intelligence (AI) applications conduct highly intensive tasks such as image recognition, language processing, and content generation. To offload AI tasks from your CPU, organizations have integrated neural processing units (NPUs) into your computer, potentially speeding up AI tasks. Learn about NPUs, how they help you complete generative AI and everyday tasks, and devices that offer neural processors.

What are neural processing units (NPUs)?

Neural processing units (NPUs), also known as neural processors, are specialized processors designed specifically for executing machine learning algorithms. NPUs are tailored to perform the massive amount of matrix and vector computations that are typical in AI algorithms. By using an architecture optimized for parallel processing, they can execute multiple operations simultaneously, which is crucial for the computational demands of deep learning models.

Think of NPUs as control centers, built specifically to handle tasks related to AI. NPUs accelerate tasks such as image and voice recognition, and natural language processing, to provide a significant performance boost over traditional CPU processing for AI-related tasks.

How do neural processors improve AI performance?

The synergy between NPUs and AI is clear—as AI becomes more sophisticated and prevalent in everyday applications, the need for specialized hardware like NPUs grows. AI applications, such as Copilot in Windows,¹ that involve complex data patterns and predictions benefit from NPUs because these processors can quickly analyze and interpret the data with high accuracy.

NPUs vs. CPUs vs. GPUs: What’s the difference?

Following the explanation of NPUs as specialized for AI tasks, it's essential to understand how they compare to central processing units (CPUs) and graphic processing units (GPUs). CPUs are the versatile core of computing, handling a broad spectrum of tasks. GPUs, originally designed for rendering graphics, are adept at parallel processing, making them useful for AI and complex mathematical calculations. NPUs, however, are tailor-made for AI, optimizing processes like image recognition and language understanding more efficiently than CPUs or GPUs.

How do NPUs help with everyday tasks?

NPUs allow smart devices to offer features like real-time language translation, enhanced photography with scene recognition, and personalized recommendations. The ability to perform these tasks locally on a machine is crucial for improving speed and efficiency. When these processes are handled directly on the device, they eliminate the latency associated with cloud-based computations and lead to accelerated results.

NPUs help localize this process and ensure that sensitive data remains more secure since it doesn't need to be transmitted to external servers. Additionally, it allows for seamless functionality even in environments with limited or no internet connectivity to enhance user experience and reliability. For users, this means faster, more responsive devices that understand and anticipate their needs better.

What devices use NPUs?

The Surface Laptop 7ᵗʰ Edition has a Qualcomm® HexagonTM NPU that’s capable of 45 trillion operations per second (TOPS). With industry-leading AI acceleration, this NPU provides unmatched experienced with lightning speed and smooth performance.

As you look toward a future intertwined with AI, the importance of NPUs cannot be overstated. Devices like Surface Laptop 7ᵗʰ Edition are just the beginning. With the ability to handle complex AI tasks efficiently, NPUs are setting the stage for a new era of intelligent and connected devices that will transform your interactions with technology. To learn more about how your computer hardware and software integrates AI, find out how you can do more with Microsoft Surface.

  • DISCLAIMER: Features and functionality subject to change. Articles are written specifically for the United States market; features, functionality, and availability may vary by region.
  • [1] Copilot in Windows (in preview) is available in select global markets and will be rolled out to additional markets over time. Copilot with commercial data protection is available at no additional cost for users with an Entra ID enabled, eligible M365 license. Your AI-powered chat for the Web with commercial data protection | Microsoft Copilot. Copilot for Microsoft 365 sold separately and requires a qualifying volume license or subscription. Microsoft Copilot for Microsoft 365 | Microsoft 365.

Products featured in this article

More articles