There's been a change in customer base for HPC, a change that has transformed the way the technology is being used. What are the underlying trends for today’s HPC solutions and what will drive future changes?

There appears to be no let-up in the rise of high-performance computing (HPC). Once the preserve of research establishments with huge budgets to spend on machines to drive excessive volumes of processing, there’s now a new generation of companies who are keen to exploit some of the changes in the HPC landscape.

What has driven this change? Greater availability of new technology, certainly. But there have been changes in demand too. For the first instance, there have been the advances in chip technology and the consequences of Moore’s Law. This has been coupled with the rise in cloud computing, which has brought high levels of processing power within reach for smaller organizations. By obviating the need for high amounts of capital expenditure on high performance hardware, HPC is affordable and accessible to a new generation of users.

The rise of cloud computing

According to Hyperion Research, "the overall spending for HPC grew from $30.6 billion (USD) in 2020 to $34.8 billion (USD) in 2021, counting both on-premises and cloud spending, an increase of 13.8%." ^[1] Hyperion Research also tracks the use of public clouds for HPC workloads, which saw an additional $5.1 billion (USD) in spending in 2021 (a rise of 18.6% compared to 2020).



But it's not only cloud computing that is enjoying a boost. According to AFCOM's 2021 State of the Data Center Report ^[2], 58% of US users are shifting deployment from public cloud to private data centers. This has been coupled with a big jump in rack density - 62% of organizations surveyed say their rack density has increased in the past three years.

These increases have been heavily driven by new levels of demand. Organizations are looking to make better use of customer data and a whole generation of companies have sprung up to support this. The market research firm IDC projected that the global AI market will reach a size of over half a trillion US dollars by 2024 ^[3].



And these companies are increasingly turning to high-density data racks – as found by the AFCOM survey. These racks are delivering the type of secure and rapid processing times that modern organizations need.

One of the biggest changes in the past few years has been the varieties of core computing on offer. For many years, the CPU was the heart of all computer processing. But GPU technology has become a vital processing partner for demanding HPC applications, enabling parallelism for complex simulations. These can include: molecular systems, climate models, and computational fluid dynamics.

Changes at the core of computing

The world has moved on even further and we've seen DPUs (data processing units) joining the HPC arsenal. AMD, for example, has acquired DPU specialist Pensando for $1.9 billion. According to Globaldata ^[4]: "the acquisition enables AMD to compete with companies in the emerging yet competitive data center market." Intel, meanwhile, has snapped up Granulate, an AI-based developer of real-time continuous optimization software. The acquisition, says Globaldata, "will help optimize and scale up Intel's cloud and data center segment."

Underpinning all these changes are the growing requirements to be energy efficient and sustainable. Data centers globally account for about 2% of CO2 emissions, but according to research from GeSI ^[5], by 2030, improved technology could mean that the emissions avoided through the use of ICT are nearly 10x greater than the emissions generated by deploying it.

Speaking of improved technology, the latest 3rd Gen Intel® Xeon® Scalable processors and 3rd Gen AMD EPYC™ processors with AMD 3D V-Cache™ come optimised for AI and the additional security demands of the modern enterprise, helping to deliver power-efficient server technology. This combination is arguably the holy grail for all businesses as they get to grips with a world that revolves around ever-growing volumes of data.