AI PCs are here: should they be on your radar?
For around a year, all the biggest processor makers have been teasing users about AI PCs. A handful of such devices have already reached the market, with many more expected to be launched later this year.
By Nvidia
For around a year, all the biggest processor makers have been teasing users about AI PCs. A handful of such devices have already reached the market, with many more expected to be launched later this year. While AI PCs are expected to boost overall PC sales, for an average user, there might not be enough compelling reasons to buy an AI PC just yet.
Generally, an AI PC can be described as a computer designed to handle AI and machine learning tasks. In addition to dedicated powerful processors (CPUs), graphic cards (GPUs), and robust RAM options, these PCs have an additional feature that is designed to boost AI tasks, known as a neural processing unit (NPU).
Its primary function is to handle AI and machine learning operations right on your PC instead of sending data to be processed on the cloud, thus enabling faster and more secure performance. NPU can enhance the efficiency of GPUs and CPUs by handling tasks such as blurring in video calls or object detection in video or photo editing, allowing a GPU to dedicate its resources to more demanding operations.
Even though GPUs and CPUs can also process workloads, NPUs are designed to do it efficiently, resulting in the longer battery life of a PC.
The performance of NPUs is measured in TOPS – Trillion Operations Per Second. Currently, the fastest NPUs from Intel and AMD have up to a few dozen TOPS.
Hardware is already here
The biggest processor makers have already released or are soon planning to release new hardware with dedicated NPUs. Intel has started shipping its Core Ultra processors, dubbed Meteor Lake. AMD announced a line of CPUs in January at CES, and this week, it released Ryzen 8000 for desktop users.
These companies were not the first to integrate processing units to handle AI tasks in their hardware. In 2017, Apple announced an A11 Bionic chip for the iPhone, which has already featured a neural engine – a type of NPU – and in 2020, a neural processor was integrated into its custom-designed M chips.
Google has been using its tensor processing units in Pixel phones designed for machine learning tasks for a few years, while Qualcomm has had NPUs in its personal computers and mobile phone processors since 2018.
This year, the company showcased a video where a PC with the Snapdragon Elite model notably surpassed a PC with an AI-dedicated Intel CPU Core Ultra. It also demonstrated benchmarks where it outperformed the MacBook with an Apple M2 processor.
The company is expected to release a few new personal computers with Snapdragon X Elite processor this year. Qualcomm says that its new processor can achieve up to 75 TOPS, the fastest on the market.
Promises for the future
The introduction of this new hardware and the buzz surrounding artificial intelligence is expected to boost overall PC sales following several years of decline. Last year, the PC market decreased by nearly 15 percent, according to Gartner. It was the second year in a row with a double-digit decline. This year, Canalys projects that overall PC sales should grow by 8 percent and around 20 percent in 2025.
Intel, for example, plans to deliver 100 million chips for AI PCs by 2025. Its CEO Pat Gelsinger said last year that he sees AI PC as a sea change moment in tech innovation. Many PC and hardware manufacturers have expressed similar views.
In the future, AI PCs promise fast, personalized, and secure AI capabilities. But right now, there don’t seem to be many compelling reasons to upgrade, at least for an average user.
From what we saw at CES, where AI PCs were one of the hottest trends, they were used to offload video-streaming tasks, improve microphone noise reduction, blur the background of a video call, or perform similar tasks that can be done without dedicated NPUs. However, a dedicated NPU can ensure smooth program operation in the background and save the device’s battery life.
Right now, we have the basic hardware for AI tasks, but there isn’t much software that can benefit users. After a year or two, the hardware will also get updated. That’s when we may see a more capable AI PC.