The end of Moore’s Law refers not to a single event but to the gradual exhaustion of the two trends that drove computing forward for decades: the steady doubling of transistors per chip, and Dennard scaling, the companion principle that let those transistors run faster without raising power density. As both slowed in the 2000s and 2010s, the easy gains that programmers had relied on for years began to disappear.
For most of the history of the microprocessor, software got faster simply by waiting for the next chip. Clock speeds climbed and single-thread performance rose generation after generation, so even unchanged code ran quicker on newer hardware. When Dennard scaling broke down around 2005, that stopped. Designers could no longer keep raising clock frequencies without exceeding power and thermal limits, and the industry’s response was to put more cores on a die rather than make one core much faster. The “free lunch” of automatic speedups was over, and extracting performance now required writing parallel software.
The historians of this shift are John Hennessy and David Patterson, whose 2018 Turing Lecture and 2019 Communications of the ACM article “A New Golden Age for Computer Architecture” laid out the consequences. They argued that with the end of Dennard scaling and the slowing of Moore’s Law, the path forward was no longer general-purpose performance from a single processor but domain-specific architectures: hardware tailored to particular workloads. As they put it, the opportunities ahead lay in “domain-specific hardware, enhanced security, open instruction sets, and agile chip development.”
This is why GPUs, originally built for graphics, became central to machine learning, and why companies designed dedicated accelerators like tensor processing units for neural networks. When you cannot make a general processor much faster, you build a specialized one that does a narrow job extremely well per watt. The center of gravity in performance moved from the clock to the architecture.
Gordon Moore himself acknowledged in his later years that the exponential could not run forever, since it would eventually collide with the size of atoms. The end of Moore’s Law is therefore better understood as a transition than a wall: the era of effortless scaling gave way to an era of architectural creativity, specialization, and parallelism, which Hennessy and Patterson framed as a new golden age for the people who design computers.