THANK YOU FOR SUBSCRIBING
The Uncertain Future of HPC
Computing has become an integral part of every aspect of society, surpassing the expectations of only a few people.

By
Apac CIOOutlook | Monday, April 24, 2023
Stay ahead of the industry with exclusive feature stories on the top companies, expert insights and the latest news delivered straight to your inbox. Subscribe today.
The future of HPC is uncertain due to the rise of cloud computing, which pressures traditional HPC vendors, but specialised user requirements may lead to dedicated HPC systems.
FREMONT, CA: Computing has become an integral part of every aspect of society, surpassing the expectations of only a few people. In the fields of science and engineering, computing is considered the third approach, alongside theory and experiment, while big data and AI are known as the fourth approach. The field of scientific computing has expanded to include data analysis, as well as disciplinary and interdisciplinary modelling. Today, exascale scientific computing systems are as costly and complex as global scientific facilities.
During the early years of digital computing, the development of supercomputers was greatly influenced by governments' funding and expertise. This had a significant impact on the subsequent mainstream and customer computing products. The present-day economic and technological power has shifted towards smartphone and cloud service companies.
Ecosystem Shifts
To comprehend the potential future of high-performance computing (HPC), it's necessary to analyse two significant changes in computing technology. The first one is the emergence of large-scale commercial clouds, and the second one is the economic and technological obstacles that have come with the development of semiconductor technology.
Cloud Innovations: Major companies bought standard servers and networking equipment to use in traditional colocation centres. As their operations expanded, they started designing specialised data centres optimised for power efficiency and developing them in locations chosen through multi-factor optimisation. Factors considered in site selection include access to cheap energy, tax breaks and political subsidies, stability of the location politically and geologically, network connectivity, and customer demand.
Semiconductor Evolution. In the past, the most consistent way to improve performance has been through advancements in semiconductors, with smaller and faster transistors and bigger and better chips. However, as chip sizes have become smaller, the pace of new technology generations has slowed while semiconductor foundry costs have skyrocketed. With the introduction of extreme ultraviolet lithography and gate-all-around FETs, maximising chip yields, minimising manufacturing costs, and achieving the best chip performance have become a more complicated task in all areas of computing, including high-performance computing.
An HPC Checkpoint
It is important to examine the current state of HPC in light of the increasing popularity of cloud services and constraints on commodity chip performance. HPC has undergone a period of significant creativity and experimentation in architecture from the 1970s to the 1990Universities and companies developed new shared memory designs, including NYU Ultracomputer, Illinois Cedar, Stanford DASH, and BBN Butterfly.
Leading-Edge HPC Futures
It is now widely accepted that supercomputing, especially at the highest levels, requires a regular investment of non-revenue capital to cover the costs of developing and deploying new technologies and integrated systems. This capital can come from other more profitable divisions of a business or external sources such as government investment.
End-to-end Hardware/Software Co-Design is Essential
The use of commonly available semiconductors has resulted in a single type of HPC dominated by processors and GPU accelerators. Although, due to current limitations in semiconductor technology, significant performance improvements will require a more deliberate and collaborative design process that spans from device physics to application development.
The Space of Leading-Edge HPC Applications is Far Broader Now Than in the Past
HPC was initially developed for solving complex optimisation problems and time-dependent, partial differential equations on intricate meshes in specific domains. While those areas will remain significant, other areas of advanced computing in science and engineering are becoming increasingly important.
Cloud Economics Have Changed the Supply-chain Ecosystem
The scale of commercial cloud infrastructure and social media company developments has surpassed even the largest HPC systems. Compared to the billions of dollars spent by cloud vendors every year, a 500 million USD supercomputer acquisition every five years provides limited financial leverage. As a result of market economics. Computing hardware and software vendors, who are relatively small compared to the large cloud vendors, now prioritise the needs of cloud vendors.
The future of high-performance computing is uncertain, with cloud computing looming large as a major player in the computing landscape. The dominance of cloud vendors and their massive infrastructure investments have put pressure on traditional HPC vendors and may drive further innovation in hardware and software. However, the specialised requirements of HPC users may also lead to the development of dedicated HPC systems that are not as heavily influenced by cloud computing. Ultimately, the future of HPC is likely to be shaped by a combination of market forces, user needs, and technological advancements.