TECH NEWS – So most PCs will have NPUs (neural processing units, or AI accelerators), which doesn’t seem like much of a surprise.
Analyst firm IDC has pointed out the obvious. That is, the proportion of computers capable of artificial intelligence will continue to grow over the next three years. That’s why the firm predicts that by 2027, 167 million PCs will be sold with hardware designed for generative, content-generating AI. Given that nearly all Intel and AMD processors and APUs (processor and integrated graphics chips) will be equipped with this technology by then, this prediction is not an exaggeration, but rather an underestimation of the potential rates. But what if AI is used to create images? An NPU may not be enough, but a sufficiently powerful graphics card (e.g. an Nvidia GeForce RTX 4060) is better. At least this is the case today, and later NPUs may evolve to be useful for this…
IDC defines an AI-enabled PC as one that has an NPU, even though GPUs are better for generative AI today. The firm divides AI-enabled PCs into three categories. The entry-level category has less than 40 TOPS (trillion operations per second) of AI processing, the next-generation machines have between 40 and 60 TOPS, and the advanced AI PC category has at least 60 TOPS.
In AMD’s latest Hawk Point APU, the NPU can do a maximum of 16 TOPS, Intel says its Meteor Lake processors can do 34 TOPS, although that includes the processor, graphics chip, and NPU in the calculation (looking at AMD’s side, their total is 39 TOPS; and the next-gen Strix Point is rumored to be capable of 45-50), and Lunar Lake has three times that, but it’s not clear whether it’s the NPU that’s getting the boost or all three components together. Apple? The M3 silicon has 18 TOPS.
So the development of NPUs could soon be accelerated significantly.