In 1965, Intel co-founder Gordon Moore famously speculated that the number of transistors on a semiconductor would double approximately every 18 months, while the manufacturing costs to produce denser, more complicated semiconductors would reduce. Interestingly, Moore was confident this relationship would be sustained over a 10-year horizon but was less certain of the two-year transistor doubling interval beyond that.

Fast forward 57 years, and it’s clear that Moore’s uncertainty post the 10-year mark was unwarranted. However, whether Moore’s Law still holds true in 2023 is a topic of debate among chip makers and computer scientists alike. Even semiconductor giants Intel and Nvidia cannot agree on whether Moore’s Law has run its course in terms of the physical limitations of shrinking transistor size.

Enter Koomey’s Law, introduced by Stanford professor Jonathan Koomey in 2010. Koomey’s Law asserts that compute energy efficiency doubles every 18 months, a trend that persisted for half a century but has decelerated since 2000 to doubling every 2.6 years.

The paradigm shift that Koomey’s Law brings – emphasising efficiency or cost per watt over compute density, may become more influential than Moore’s Law in the decades to come. This shift has significant implications not only environmentally and economically, but also in reshaping how we design digital experiences and consume digital products.

There is practical evidence of this shift. Amazon Web Services introduced the first of its custom ARM-based Graviton processors in 2018, releasing the Graviton2 processors two years later, with the promise of a price performance improvement of up to 40% over equivalent Intel and AMD x86 CPUs, and the Graviton3 last year, which makes a 25% price performance improvement on Graviton2 chips.

Aligning with the industry’s shift towards more sustainable and cost-effective computing solutions, Microsoft partnered with ARM processor developers Ampere in launching its Azure ARM instances in 2022 and has recently announced the intended 2024 release of its first in-house designed microprocessor for cloud computing: the Azure Cobalt 100, a 128-core, 64-bit ARM-based processor with a touted 40% reduction in power consumption over existing ARM-based instances available within Azure.

ARM in the data centre

ARM microprocessors, renowned for their energy efficiency and low power consumption, are widely used in mobile and edge computing. The trend towards ARM-based silicon in end-user devices has been notably advanced by Apple with its M1 chip in 2020, and recently the release of its 3-nanometer M3 chip, which offers up to 50% improved performance over the M1 through a combination of performance and efficiency cores. This development may give rise to a potential shift among other device manufacturers in releasing their own ARM-powered laptops as workhorse daily drivers.

This brings me to where I see ARM’s biggest potential for impact: the enterprise data centre. The start of my career in the early 2000s coincided with the burgeoning adoption of ‘Lintel’ architectures in the enterprise, driven by the affordability of Intel x86 processors and the rise of enterprise-grade Linux distributions such as Red Hat and SUSE, and offering cost-effective alternatives to pricier, proprietary Unix variants. In fact, some of the earliest projects I worked on involved migrating applications from DEC Alpha Tru64 and Sun Solaris Sparc-based systems to Red Hat environments running on cheaper Intel x86 hardware. In short, I feel like I’ve seen this movie before.

At LSD Open, I have the (awesome) opportunity to talk to technical leaders across a variety of verticals, and frequently their challenges and objectives are consistent: the cost of mitigating the impact of load shedding and stockpiling diesel to ensure service continuity in the event of an extended blackout — all of which have a significant impact on operating margins.

Then there’s the need to modernise and automate to drive scalability, efficiency and sustainability, not just at the digital experience layer but all the way down to the data centre and infrastructure level — leveraging “cloud-like” data centre orchestration, where the addition of compute, storage or network capacity to cater to surging demand is available automatically and driven by machine learning and advanced telemetry.

Excitingly, these capabilities exist. At LSD Open, we have been building edge computing solutions for our customers, combining the portability of containerisation with the orchestration and self-healing capabilities of Kubernetes, all running on robust, low-cost and energy-efficient ARM processors, can and should make their way into the data centre – and this applies doubly to the South African data centre.

This is not to downplay the complexities in both changing CPU architectures or adopting cloud native as an operating model. However, the advantages – and savings – well outweigh any perceived risks. Most modern languages have mature multi-architecture support and often you’re only a compiler flag away from producing an ARM-compatible build.

While ARM-based CPUs may not currently reach the clock speeds of their Intel and AMD counterparts, Kubernetes, coupled with scale-out microservices application architecture and pluggable “blade-like” ARM boards means that adding compute and elastically scaling workloads is simple, automated, and generally more cost-effective and flexible than vertically scaling workloads.

While I’m not saying that ARM will bring about the death knell of the x86 server, or topple semiconductor giants such as Intel, I do believe that when combined with cloud-native architecture, it brings a very real and important value proposition not just to the data centre and the organisations which operate them, but more broadly to a new way of designing digital experiences and the applications which underpin them, one which considers sustainability as a first principle — and who doesn’t want to live in that digital future?