TRADE CHAT: Are cheap chips a thing of the past?

I'M NOT JUST TALKING ABOUT OVERPRICED graphics cards, but the underlying cost of manufacturing advanced integrated circuits. A chip manufacturing expert from Google, Milind Shah, thinks so, recently declaring that the per-transistor cost stalled as far back as the 28nm node over a decade ago. That's pretty worrying, but it can't possibly be true. Here's why.

- Jeremy Laird

This gloomy observation has been accompanied by data plotting the cost of 100 million transistor gates across multiple generations of chip production. It shows the cost plummeting from 90nm to 65mm, falling steeply again to 45nm, with another decent drop at the 28nm node. And then nothing.

Actually, it's worse than that, because the graph shows a gentle incline. It's more expensive to manufacture 100 million gates using today's most advanced 5nm and 3nm nodes than 28nm silicon.

But this can't be true. Let's take Nvidia's 28nm GK104 GPU. It first appeared in 2012, and was used in various GeForce graphics cards from the GTX 680 downwards. GK104 contained 3.5 billion transistors. Today, the current Nvidia 104-class chip, AD104, as found in the RTX 4070 and 4070 Ti, clocks in at 35 billion transistors.

That's 10 times the number of transistors. Is Nvidia really paying 10 times as much for an AD104 die today as it did for a GK104 die in 2012? Graphics cards have gotten pricier, but not by that much.