There's a lot of talk these days about NPUs, especially since the launch of Copilot+PC, the software and hardware suite introduced by Microsoft on May 20th.
An NPU is a processing unit designed to handle the low-precision algorithms required by artificial intelligence. They do this more efficiently than CPUs because they are specifically designed for this purpose. A CPU, on the other hand, is designed to handle various types of algorithms, which means they consume more energy and take more time because they are not specialized.
Are NPUs a new type of chip?
They are neither a new type of chip nor separate from others. On the contrary, they are integrated within CPUs.
Which CPUs have NPUs?
- AMD Phoenix and Hawkpoint (nomenclatures 7x4x and 8x4x), like the 8840HS with up to 38 TOPS (16 of NPU).
- Intel Meteor Lake (nomenclatures 5/71x5H), like the 7155H with up to 34 TOPS (11 of NPU).
- Snapdragon X Elite, up to 45 TOPs.
Are we going to integrate CPUs with NPUs?
Actually, we already do. The AMD7000 Excalibur already has them, and in a few weeks, they will be released with the AMD8000 series. Our roadmap includes integrating more options, much to the delight of AI enthusiasts.
Is an NPU better than a GPU (graphics card)?
Current NPUs can reach 40/45 TOPS, while NVIDIA's RTX graphics cards can range from 130 to over 1000 TOPS. However, graphics cards generate a lot of heat and consume a lot of energy—so much that your computer’s battery life can be reduced by 6 or 7 times.
Can I run AI programs without an NPU?
Yes, in fact, here you can see that there are already people who have installed Copilot+PC on a computer without an NPU, but the processing is slower. And we’re not talking about a computer with a dedicated GPU, as it didn't have one either.
This is for Windows, but what about Linux?
Indeed, the use of artificial intelligence as a major aid in your operating system is also being developed for Linux, as noted on Intel’s official website.
Should I buy a computer now or wait until 2025?
If your main goal is to produce AI, I recommend getting a computer with an NVIDIA graphics card. NVIDIA GPUs are renowned for their superior performance in deep learning tasks and intensive data processing.
On the other hand, if you can wait until 2025, you will likely benefit from the new technological advancements currently being developed. Both AMD and Intel are expected to release new generations of processors that offer even better performance and energy efficiency. Additionally, specialized AI hardware, such as Neural Processing Units (NPUs) and improvements in GPUs, will continue to advance, providing more powerful and efficient tools for AI consumption and production.
Conclusions:
From all the above, we consider NPUs to be the present and future of computing. We see them as useful units for consuming and using AI, but they fall short for AI production, for which a GPU is still better.
If you want to stay up to date with our news, don’t forget to subscribe to our newsletter (we promise not to spam you).
Read more of 2025 launches, in
WCCFTech.
