Hardware enables software; therefore, hardware enables digital business. Without hardware, neither can succeed.
Utilizing the right hardware should be just as important to an organization as the platform and network they choose. Why? Because performance continues to remain supreme, and the performance of digital business hinges on utilizing the best technology to meet demand. Selecting appropriate hardware enables efficiency in cost, in power to operate, and in agility.
Today’s xPU technology is the enabler for economies of scale because it exponentially improves performance and drives digital transformation through a specialized piece of hardware that contains specific acceleration functions in silicon that are meaningful to networking and applications. But to leverage the maximized computing and processing power of xPUs, organizations need to modernize their enterprise architecture around their applications.
If you’re reading this, I’d like to assume you’ve heard of Moore’s Law. However, in the event you haven’t, it should be clear that Moore’s Law isn’t a law, it’s more of a prediction—one that happened to ring true for roughly 50 years. In 1965, Gordon Moore of Intel calculated that the number of transistors per integrated circuit—a CPU—would double every year as technology advanced, delivering exponential improvements to processing power. He later amended this to be a doubling every two years, but over time the numbers worked out more closely to every 18 months.
Over the past decade or so there’s been a slowing in the CPU improvements produced by hardware updates, which has led to many claims that Moore’s Law is dead or dying. But those claims are based too directly on the doubling of components on the raw silicon and improvement of general processing.
It’s true we are approaching a point at which hardware components can’t get any smaller—thanks to the underlying physics—but that doesn’t mean computing and processing improvements are stagnating. The remedy to our CPU obstacle is specialized hardware—xPU—designed to handle specific software needs.
Think of a CPU like your standard chef knife. It’s versatile and capable of doing all the work needed to cook and serve a meal, albeit with some inefficiencies. Specialized compute—the GPU, DPU, and xPUs of today—is equivalent to the precision pieces found in your knife block, such as a filet knife, meat cleaver, and kitchen shears. When utilized in addition to CPUs, they provide greater speed and efficiency in the process of preparing and serving a meal, enabling the cook to “scale out” their offerings.
While we may not continue to see a doubling of transistors and general CPU power every 18 months past certain physical boundaries, precision or “domain-specific” hardware’s performance capabilities continue to enable exponential improvements. This effectively extends Moore’s Law past general purpose compute, provided the software and architecture are designed to take advantage of the accelerated functions available inside an xPU’s capabilities.
In the past, the doubling of transistors and compute power inside of general-purpose CPUs was enough to adequately accelerate performance. Today, however, it’s the pairing of precision hardware with software designed to leverage its special computing capabilities that improves the operational experience.
Designing an application with CPU in mind means developers need to be cautious about data needs as it will affect the speed, quality, and performance of the application. However, creating an application with a DPU in mind means the application can be designed to take advantage of faster data processing, while simultaneously using less power. Referring to our earlier analogy, it’s like a chef using a meat cleaver to prepare a chop instead of a standard chef knife; they save time and energy.
But inserting a top-of-the-line DPU into a data center won’t simultaneously optimize data processing and enable observability. The enterprise architecture must be designed to maximize the hardware’s capabilities for the software it hosts, meaning hardware itself must be thought of as an essential architectural component.
Watch a TV show or movie from the 90’s on today’s 4K UHD TVs and you’ll have a visual representation of software not designed to take advantage of the hardware. And trying to place one of today’s 4K processors in an old CRT or rear-projection TV won’t give viewers the upgraded experience they desire since the system wasn’t architected to capitalize on such technology. Only the proper pairing will give you the best visual. Similarly, current enterprise architectures weren’t developed to leverage the capabilities of today’s xPU hardware. Organizations are—today—preparing to launch their digital transformation journey with modern software atop the hardware of the past; this can only lead to future failure.
To learn how to modernize architecture to capitalize on today’s hardware capabilities and enable a digital business at scale, read “An Infrastructure Renaissance,” a chapter by CTO of Systems and distinguished engineer Joel Moses in our new O’Reilly book, Enterprise Architecture for Digital Business.