Moore's law has slowed somewhat for CPUs, and Intel announced a few years ago that they were intentionally slowing down the pace of their releases.

"Intel stated in 2015 that the pace of advancement has slowed, starting at the 22 nm feature width around 2012, and continuing at 14 nm.[20] Brian Krzanich, CEO of Intel, announced, 'Our cadence today is closer to two and a half years than two.'"

https://en.wikipedia.org/wiki/Moore%27s_law

"Intel disclosed in a regulatory filing that it is slowing the pace with which it launches new chip-making technology."

https://www.technologyreview.com/s/601102/intel-puts-the-bra...

Moore's law for GPUs, on the other hand, hasn't slowed down yet because it's not as close to the wall.

https://blog.openai.com/ai-and-compute/

https://www.pcgamer.com/nvidia-ceo-says-moores-law-is-dead-a...

(Tremendous hubris warning, please correct me if I'm off base with where I go)

Moore's law doesn't include capital investment as a factor.

Initially, it could be mostly ignored as it was swamped by available profits (at least for microprocessors, less so for memory). More recently, I believe it's been matched by increased volume of semiconductors used throughout the global economy.

But at some point we'll reach saturation. And costs to shrink feature size will continue to increase exponentially. So it simply won't make financial sense to continue.

(Unless non-silicon, non-electrical, or radical patterning technology hits a breakthrough)

Moores law held up pretty well. Frequency scaling did not continue as it did in the old days, I mean we had 3ghz back in the P4 days. But performance has certainly improved thanks to better IPC, more cores etc.

I remember reading about IBM Power chips running at 5GHz years ago (maybe it was https://hardware.slashdot.org/story/07/01/04/1824223/ibms-ne... ).

Presumably they were too expensive (manufacturing, power, cooling, etc.) since they're not ubiquitous. IIRC the "barrier" to going any higher was the speed of light across the chip: running faster than 5GHz would require the output to be causally independent from the input!

The law is about transistors I believe, not about frequency or performance, isn’t it?

as originally formulated, yes, but no one really cares about transistor density itself; we care because it correlates well with performance.

Not quite 5ghz. We have P8 based E880Cs running SAP @ 4.95ghz, iirc.

Both.