the post-Moore era
/ MOORE rhymes with 'sure' /
For about forty years, computers had a magic gift that came almost for free. Every couple of years, the same chip could be redrawn with smaller transistors, and you got roughly twice as many of them in the same space, each one a little faster and using a little less power. Programmers did nothing and their old code ran faster on next year's machine. It felt like a law of nature. The 'post-Moore era' is the name for the world after that gift faded — where transistors keep shrinking slowly but the free speedups have stopped, and architects must work for every gain rather than wait for it.
Two separate things ended, and it helps to keep them apart. Moore's law was an observation about transistor count: the number of transistors you can cheaply pack onto a chip kept doubling roughly every two years. That trend has slowed and grown enormously more expensive, but it has not fully stopped. The deeper break was Dennard scaling, the rule that as transistors got smaller their power density stayed constant, so you could run them faster at the same total power. Dennard scaling effectively ended around 2005: shrinking transistors no longer kept power in check, so clock speeds stalled near a few gigahertz and could not simply keep climbing. That is the real wall, and it is why your laptop's clock rate is not ten times higher than one from 2005.
The whole rest of this field is the answer to one question: if a single general-purpose core can no longer be made dramatically faster each year, where do future gains come from? The broad consensus is three places. Specialization — building hardware tuned to one job, like a chip for machine learning, instead of one chip that does everything adequately. Better memory systems — because moving data, not computing on it, is now the dominant cost. And software-hardware co-design — designing the chip and the program that runs on it together. The post-Moore era is not the end of progress; it is the end of progress you got for free, and the beginning of progress you have to design.
Around 2004, a single desktop core might run at 3 GHz. Twenty years later, top cores still run at roughly 3 to 5 GHz — not 30 GHz. Instead of one core ten times faster, a 2024 chip might have 16 cores, several specialized accelerators, and a faster memory system. The transistors got smaller; the free clock-speed escalator stopped.
Transistors kept shrinking, but clock speed plateaued near 2005 — the visible signature of the end of Dennard scaling.
Do not conflate the two laws. Moore's law (transistor count) is slowing but not dead; Dennard scaling (performance-per-watt at fixed power) effectively ended around 2005. The death of Dennard scaling — not Moore's law — is what forced multicore and accelerators.