When the average person thinks about high-performance computing (HPC), they often imagine government or university labs doing advanced scientific research. They might imagine astronomers analyzing the radiation emitted by quasars, geneticists unlocking the secrets of the human genome, or physicists unraveling the mysteries of what happens when atomic particles collide.
Of course, many researchers are using HPC systems for projects exactly like these. But the most common use of HPC technology is quite different.
A recent Intersect360 Research report reveals that about half of all HPC spending comes from commercial users. Manufacturers, pharmaceutical companies, oil and gas companies, and businesses in a host of other industries are some of the biggest users of HPC.
And businesses aren’t just purchasing a few HPC systems, they are increasing the rate at which they purchase them every year. By 2025, Intersect360 predicts that worldwide HPC spending will top $60 billion.
So what are businesses doing with all that high-powered infrastructure?
To a large extent, the answer is analytics. While basic analytic techniques like data mining have been around for decades, the more recent trends toward artificial intelligence (AI) and machine learning (ML), are driving more companies to invest in HPC.
Machine learning is a subset of artificial intelligence, and it is a fairly new use case for HPC. According to the Intersect360 report, “Traditionally, most HPC applications have been deterministic; given a set of inputs, the computer program performs calculations to determine an answer. Machine learning represents another type of applications that is experiential; the application makes predictions about new or current data based on patterns seen in the past.”
Today, a wide range different types of applications include machine learning capabilities. Machine learning tells factories when their machines are likely to break down. It figures out how world events are likely to affect the supply chain. It tells oil companies where to drill. It alerts credit card companies when someone might have committed fraud. It assists hospitals in diagnosing their patients’ illnesses. It helps marketers figure out how to segment their customer base to maximize their profits. And it tells drug companies how different chemicals will affect the human body.
In fact, 81 percent of HPC users surveyed in 2021 said that they were already running some machine learning workloads in their environments or they planned to do so within a year.
Activities closely related to analytics and machine learning were also very common among HPC users. Intersect360 found that 43 percent of organizations use their HPC systems for visualizations, and 31 percent use HPC to run business analytics.
The Need for Greater Scalability
As the workloads running on HPC systems has changed so has the type of hardware that organizations need to support their applications.
HPC systems have always needed plenty of processor cores, lots of RAM, and large amounts of storage — and that is still the case. But today’s HPC systems are far more likely to make use of accelerators to speed up processing. The research found that 91 percent of HPC users have systems with accelerators. It is particularly common for systems to make use of graphics processing unites (GPUs), which are very effective at processing machine learning workloads.
Most organizations investing in HPC are looking for scalable, flexible infrastructure that has been tailored for analytics, artificial intelligence, and/or machine learning. Often they turn to converged or hyperconverged systems that make it easy to build a large cluster that provides exceptionally fast performance.
The report also notes that a couple of vendors have become particularly popular with organizations purchasing HPC systems that will be used for analytics. Dell Technologies is the industry leader in HPC revenue, and it was the vendor most often cited in a 2021 HPC user survey.
Many of the Dell HPC systems rely on AMD EPYC™ processors. Intersect360 found that 70 percent of all HPC users have systems with at least some EPYC processors, and 23 percent use the EPYC processors broadly. AMD’s Infinity Fabric™ allows it to speed up the flow of data between CPUs and GPUs and other accelerators, allowing for extremely fast performance, which is ideal for analytics applications.
Because companies are using their HPC systems to drive real value, this enhanced performance offers a competitive advantage to businesses in a wide variety of industries.
The report concludes, “The new HPC is fueled by analytics and AI.” It adds, “Across domains, there is a deep wealth of data. If harnessed, it can unlock new discoveries, which themselves will spawn new areas of research. While the fundamental drivers of research remain unchanged, these converged, high-performance solutions are critical to enabling new generations of insight.”