[jguimont]

> Intel 7nm is estimated to deliver around 200 to 250 million transistors per square millimetre

> TSMC reckons its 3nm node will pack in somewhere north of 250 million transistors per square millimetre of silicon

Why is a 3nm process would yield the same transistor density as a 7nm one? Is it an error in the article or there is something I do not understand?

[sroussey]

It used to be based on a feature size of logic based transistors, but as node shrinks happened the design of transistors changed, so the number didn’t make sense any more.

Different companies measure different things, and it is a marketing designation now. Intel has tried to be more scientific about their naming, but see where that got them.

[llampx]

Intel could call it whatever they want but they've been stagnating, and that's why they are failing. Its not even that they're stagnating that's the problem, its that the rest of the industry isn't.

[api]

Clock speeds became similar by the 90s at least since things like instructions per clock and instruction level parallelism can make lower clock speed chips faster than “faster” chips on most code.

[raxxorrax]

Ridiculous really that we now have it as a metric. Transistor count is more sensible, but even that is just a very indirect and approximative measure of performance.

But yes, it is shown again that they should have lied. Just don't do it too much on sensitive topics like emissions...

There probably isn't that much room for improvement anymore since we reach atomic magnitudes.

[distances]

How has that been shown? Intel's problem isn't the process naming but the lack of results.

[cultus]

From what I understand, it is usually the minimum size rather than any kind of average.