What’s going on with the chiplet market?

&xcust=2-1-2303075-1-0-0&sref=https://www.pcworld.com/feed" rel="nofollow">Like the McDLT,” says Will, “so the hot stays hot and the cool stays cool.”

Now chip makers like AMD and Intel are applying this idea to the CPU itself, isolating disparate components to different parts of the die, like the much-vaunted NPUs that are making everything an “AI PC” this year. The platonic ideal of this concept is an open standard, where you can have primary compute cores from one company with an integrated GPU from another company and a memory controller from another, all working harmoniously on a single standard.

That…isn’t what’s happening. Instead AMD, Intel, Nvidia, and the rest of the industry are all offering compelling combinations of chiplets on their own standards. These are increasing performance, efficiency, and flexibility, but we’re nowhere near the point where a major industry player can order a custom-designed processor to fit its needs.

And the engineering and manufacturing costs that go into this technology, especially layering chiplets in “3D” stacks, makes them both expensive and slow to iterate. Plus, companies aren’t exactly keen to share that expensive development with anyone else, so we’re seeing a lot of proprietary closed walls around these unique designs.

“Is Moore’s Law dead?” asks Will, invoking a frequent talking point of the tech industry.

“No,” counters Ian, “Don’t believe everything Jensen [Huang, Nvidia CEO] says.” We thought we’d reached physical limits for processor innovation before, and blown past them every time. “If Moore’s Law wasn’t still a thing, the semiconductor industry would stall. And right now we’re in the fastest growing pace of the semiconductor industry we’ve ever had.”

For more insights on the industry from some of the brightest minds in tech, be sure to subscribe to PCWorld on YouTube.

https://www.pcworld.com/article/2303075/whats-going-on-with-the-chiplet-market.html