Moore’s Law’s swan song and its impact

Moore’s Law (not to be mistaken with Murphy’s Law), proposed by Gordon Moore in 1965, initially observed that the number of transistors on a computer chip doubles approximately every two years. Over several decades, this observation held true, and it became a guiding principle for the technology industry. However, it is no longer as accurate and can be seen as almost obsolete for several reasons; let us explain.

• Physical limits: size matters

As transistors on microchips became smaller and smaller, they approached the physical limits of what was possible at the atomic scale. At a certain point, it becomes incredibly challenging to make transistors any smaller without encountering quantum effects and other physical constraints. Stephen Hawking, the cosmologist, once pointed out during a visit to Silicon Valley that nothing can travel faster than the speed of light and that materials are ultimately made of atoms of finite size. As he recalled, there are, in other words, speed and size limits to chips. (Moore’s Law: further progress will push hard on the boundaries of physics and economics)

• Economics: tiny chips, massive cost

Shrinking transistors requires increasingly expensive fabrication processes, including specialized equipment and facilities, which has made it economically unsustainable to continue doubling transistor density at the same pace, especially when the companies will need so much time to recoup their investment with little to no visibility on the semiconductor market state in 20 years+. “With TSMC’s newest factory costing $33bn – far more than the $15 to 20bn of 5 nm plants – sustaining Moore’s Law is a game of very high stakes. In this rarefied atmosphere, only a handful of players – IBM, Intel, Samsung, and TSMC – are capable of developing next-generation semiconductor chip technology. They certainly haven’t given up on Moore’s Law, but further progress is going to be very hard to sustain.” (Moore’s Law: further progress will push hard on the boundaries of physics and economics)

• Heat dissipation: not cool at all

Smaller transistors generate more heat, which can be challenging to dissipate effectively. Excessive heat can lead to reduced performance and reliability, requiring innovative cooling solutions, such as thermal transistors (Test Processor With New Thermal Transistors Cools Chip Without Moving Parts | Tom’s Hardware)

• Energy efficiency

As transistors become smaller and more densely packed, energy efficiency is becoming a significant concern. Power consumption is increasing as more transistors are packed onto chips, leading to higher energy costs and environmental concerns. (Forget Moore’s Law—Chipmakers Are More Worried About Heat and Power Issues – IEEE Spectrum)

• Shift in focus: no R&D, no party

The technology industry has shifted its focus from simply increasing transistor count to improving other aspects of computing, such as parallel processing, graphics, and specialized hardware for tasks like artificial intelligence and machine learning.

• New architectures: futurism is here

Advances in computing have led to the development of new architectures, such as multi-core processors, GPUs, and TPUs, which offer performance improvements in different ways.

While it’s still possible that researchers will find a way to keep minimizing transistors and work around all the challenges that it entails, the data in recent years shows that the rate of increase in processing power is definitely slowing down.So, how does this affect your company’s IT infrastructure needs? We dare to give you our predictions.

• We need more data centers

The digital service demand is already growing at a much faster rate than the processing power – even back when Moore’s Law was more applicable, which means the companies already had to increase their data centers’ capacity, even though the processing power was growing at a fast rate. In the future, this means even more servers and data centers will be needed, with all the sustainability challenges that this will bring. You might not like these big black boxes being built around the corner, but at this stage, they are an evil necessity (unless we all stop using the internet at once, that goes without saying).

• We should embrace sustainable practices all together

Companies will have to find new solutions to tackle the sustainability challenges. Energy consumption will rise significantly, especially given that the number of transistors per chip will remain relatively similar.

We keep repeating ourselves, but it’s because we really believe in it: using circular hardware and keeping the equipment alive longer will make more and more sense as the CPUs’ capacity increases less and less. It will provide substantial savings, both for companies and the planet.

• The chip market will become way more competitive

For decades, AMD and Intel have dominated the market, and competitors had minimal incentive to enter it as long as the processing power was steadily rising. Now, it makes more sense to develop highly specialized chips that will attain better performance in particular areas. It can be exciting since, according to the Advanced Electronics Practice at McKinsey & Co., the semiconductor market follows a winner-takes-all model: “If a company’s product or service is even slightly better than a competitor’s, it typically captures an outsize portion—or even the vast majority—of industry revenue.”

• Workload optimization is a must

Minimizing workload CPU usage has traditionally been a wise choice for companies aiming to cut hosting expenses. It will become even more crucial now, and we will probably see more workloads move to containers, for example. FinOps and the cloud cost optimization market will probably gain in importance, too. (What Does the End of Moore’s Law Mean for the Data Center Industry?)

To sum up, we can’t hold a funeral for Moore’s Law yet, but the significant challenges and investments that come with doubling the processing power every two years are certainly marking it less relevant, and we will most likely encounter the physical limits in a not-so-distant future. This will mean that you will need to adapt and optimize existing and future infrastructure, either by seeking new highly specialized chips tailored to your needs, reducing energy consumption, improving workload processes, or repurposing any infrastructure that no longer suits your needs.

Whether you need to virtualize, implement AI, or invent a new solution, we know our chips and, Moore’s Law or not, we are always happy to help!

More general sources: https://spectrum.ieee.org/power-problems-might-drive-chip-specialization

https://www.csis.org/analysis/moores-law-and-its-practical-implications