Products You May Like
By now, we’ve all heard the term “AI PC.” This new breed of computer is reportedly coming to market soon to allow AI performance far beyond what our current computers are capable of. But what exactly constitutes an actual AI PC? Is it the presence of a specific CPU, GPU, or a combination of the two? Intel has now answered that question by laying out what hardware is required for a PC to be labeled an AI machine. Intel also stated that it will not make it a separate brand for the company as it does with some of its current processors and platforms.
The name Robert Hallock might ring a bell if you’ve been around the PC world for a while, as he used to be one of the on-camera guys at AMD when discussing that company’s CPU technologies. He’s now working for Intel and has addressed the future of AI PCs with The Register. He stated that for a PC to receive the coveted “AI” label, it needs four things: a neural processing unit, a GPU, and the ability to handle Vector Neural Network Instructions (VNNI) and DP4a instructions (so that the GPU can handle video processing). Not surprisingly, those elements exist in the company’s current Meteor Lake CPUs and are expected to be included in its CPU portfolio.
Intel’s Meteor Lake CPUs are the company’s first to feature neural processing unit, or NPU, for AI applications.
Credit: Intel
Given all the terminology involved, an AI PC is just a computer with a new CPU, so Hallock says the company won’t be branding AI PC as a platform like it does with vPro or Centrino. The Register cheekily likens it to Intel, saying an AI PC is one with its latest CPU, of course. Therefore, the phrase “AI PC” will not denote a specific type of security or feature set per se. Instead, it’ll be a computer that can run certain tasks more efficiently, most of which don’t currently exist but will someday.
“The reason we have not created a category for it like Centrino is we believe this is simply what a PC will be like in four or five years time,” says Hallock. He added that Intel will not be setting system requirements for what constitutes an AI PC regarding memory size, I/O speed, or storage capacity. That’s because these futuristic AI applications, including running LLMs locally on-device, will vary in size and storage and processing power requirements. At the same time, existing PC will still be able to run AI applications, says Hallock, it will just be extremely slow. Notably, Microsoft has reportedly determined an AI PC will have a minimum requirement of 16GB of memory.
This compares with how we used to run games without GPUs, relying on the CPU instead. Hallock says this is analogous to what an NPU can do for AI tasks and that in a few years, they will be just as ubiquitous as GPUs are today, and we won’t even think about running anything related to AI on our CPUs.