Week 49: Was 2024 the end of Moore’s law?
In 1965, Gordon E. Moore, co-founder of Intel, observed that the number of components on a single chip doubled roughly every two years at minimal cost. This observation, known as Moore’s Law, has guided the semiconductor industry for decades. Although not a scientific principle, Moore’s Law became a reliable benchmark for predicting technological progress and exponential growth in microprocessor performance.
However, in 2024, the industry faced significant challenges in adhering to Moore’s Law. Physical limitations, rising costs, and increasing complexity of manufacturing have made it harder to maintain the pace of progress.
Moore’s Law Challenged
NVIDIA
At NVIDIA’s GTC 2024, CEO Jensen Huang announced the Blackwell GPU, a groundbreaking product with 208 billion transistors capable of processing trillion-parameter AI models 30 times faster than previous generations. Huang declared this as the end of Moore’s Law, citing NVIDIA’s own achievements—a thousandfold increase in computational power over eight years—as evidence that industry advancements have surpassed Moore’s benchmarks. He introduced the term Hyper Moore’s Law to describe this accelerated pace of innovation.
Despite these advancements, Huang acknowledged that growing industry demands continue to outpace progress, signalling a new era in computing.
AMD
AMD’s CTO, Mark Papermaster, expressed a different viewpoint, suggesting that Moore’s Law remains relevant but with caveats. Papermaster estimated it could hold for another six to eight years, though the traditional doubling of transistor density every 18 to 24 months is no longer sustainable under the same cost constraints. Instead, AMD is shifting its focus to innovations in chiplet architecture, describing this approach as a Moore’s Law equivalent that can sustain performance improvements.
Intel
Intel has openly acknowledged the hurdles posed by approaching physical limits as transistors near-atomic scales. The company has faced delays in advancing its 10nm and 7nm processes, raising doubts about its ability to uphold Moore’s Law. While Intel remains committed to transistor scaling, it has conceded that traditional methods of doubling transistor counts are no longer practical. This shift reflects broader industry trends away from strict adherence to Moore’s Law.
Conclusion
The challenges to Moore’s Law highlight the ingenuity of human innovation. Historically, breakthroughs often defy pessimistic predictions. For example, early efforts in machine translation were deemed nearly hopeless in the 1960s, yet today, smartphones offer free translation apps supporting over 100 languages —demonstrating exponential leaps in computational power.