The first laptops with Windows 11 and Copilot+ are practically here. Acer, ASUS, Dell, HP, Lenovo, and Samsung are among the manufacturers joining Microsoft and Qualcomm’s initiative. This partnership relies on two key components: Copilot+ (Microsoft’s AI integrated into Windows 11), and Qualcomm’s Snapdragon X Elite and Snapdragon X Plus chips.
Microsoft chose Qualcomm’s Snapdragon X Series chips because these SoCs include a very special component: a Neural Processing Unit (NPU). While Intel’s Core Ultra processors with Meteor Lake microarchitecture also have an NPU, they lack a crucial feature that the Snapdragon X Elite and Plus chips do offer: the ability to deliver a performance of up to 45 tera operations per second (TOPS).
The Snapdragon X Elite and Plus SoCs at a Glance
These Qualcomm chips, which will power the first laptops with Windows 11 and Copilot+, boast ARM implementations with an Oryon CPU, an Adreno GPU, and a Hexagon NPU. The Snapdragon X Elite SoC operates at a maximum clock frequency of 4.3 GHz and supports a main memory map of up to 64 GB. In addition, Qualcomm claims that it offers up to 22 hours of battery life–although it's better to take such claims with a grain of salt when they come from the manufacturers.
In this article, we’re focusing on the NPU of these new Qualcomm SoCs, as it's the part responsible for executing most processes related to Copilot+. Microsoft requires SoC manufacturers to ensure their NPUs deliver a minimum performance of 40 TOPS, and the Snapdragon X Elite and Plus SoCs, as stated above, meet this requirement–and then some.
NPUs are designed to handle AI algorithms, particularly generative AI, as efficiently as possible. Currently, Qualcomm, Intel, AMD, Apple, Samsung, and Huawei are some of the companies that have chips that can deal with an approach known as heterogeneous computing.
These chips bring together three different types of processing units: the CPU, whose purpose is general; the GPU, which is primarily responsible for dealing with the processes required to generate graphics; and, finally, the NPU, which deals with AI processes. The heterogeneous architecture distributes the responsibility for the execution of processes among three different types of processing units with the purpose of maximizing performance and energy efficiency.
This summarizes the essence of heterogeneous computing. However, it’s crucial to note that AI processes can run on the CPU, the GPU or the NPU, with each choice bearing consequences.
The GPU excels in massively parallel processing, making it highly effective for graphics generation and beneficial for AI tasks, particularly if it incorporates specific units for matrix operations that support parallelization. Although GPUs deliver the highest performance for AI processes, they are also less energy-efficient.
The CPU can also execute AI processes, but with a lower performance compared to the GPU, due to its general-purpose architecture. It's also supposed to consume less energy. Lastly, the NPU is specifically designed to deal with AI processes as efficiently as possible in terms of energy, although with a lower performance than the CPU and, especially, the GPU.
Now, back to the Snapdragon X Elite and Plus SoCs. At its presentation, Microsoft boasted that these Qualcomm chips, found in the Surface Pro and the Surface Laptop, can outperform Apple’s M3 SoC by up to 23% in multithreaded scenarios. However, it’s worth noting that while the M3 chip features passive cooling, Qualcomm’s SoCs require active cooling–that is, a fan. (On a side note, at minute 3:15 of this video published by The Wall Street Journal, Microsoft CEO Satya Nadella claims that, so far, he has never heard the fan of his Surface laptop).
Intel and AMD Gear Up to Tackle the New Copilot+ PCs
Intel and AMD have reacted more slowly than other SoC and APU designers, but both companies now have chips that include an NPU, and in the future they will extend this strategy to other product lines. As mentioned earlier, Intel Core Ultra processors with Meteor Lake microarchitecture include an NPU, and AMD’s Ryzen 7040 and 8040 have an NPU with XDNA architecture optimized for AI and signal processing. The problem is that none of these SoCs can handle Copilot+ on Windows 11, as their NPU performance is well below the 40 TOPS that Microsoft requires.
However, both companies are currently putting the final touches on new SoCs that are capable of working with Copilot+. Intel will launch its Lunar Lake SoCs during the third quarter of this year, and even though we still don’t know the specs, we do know something important: the NPU in these chips will deliver more than 40 TOPS, and the GPU will exceed 60 TOPS. Intel has confirmed that more than 20 laptop manufacturers are interested in its Lunar Lake SoCs: Starting in the third quarter of 2024, more than 80 laptops with Windows 11 and Copilot+, and equipped with Intel SoCs, will begin to arrive in stores.
We also know that AMD is working on its new Strix Point SoCs, which, like Intel’s Lunar Lake, will include an XDNA2 NPU with a performance of at least 40 TOPS to be able to handle Copilot+. AMD hasn’t released too much official information regarding these chips, but some interesting facts have leaked.
The Strix Point chips will presumably feature an advanced RDNA 3+ GPU capable of competing with the high-end dedicated GPUs currently available from NVIDIA and AMD, as well as a Zen 5 CPU. If this is confirmed, it will be a big deal. The company hasn’t announced when these chips will be ready, but Lisa Su, president and CEO of AMD, hinted at an event in China in late March that they will arrive before the end of 2024. In other words, we can start getting excited.
More Information | Qualcomm | Intel
Related | Microsoft Introduces Copilot+, the Next Generation of AI-Powered PCs