scorecard
  1. Home
  2. sustainability
  3. article
  4. Supercomputers are faster and more powerful — but need to be more energy-efficient

Supercomputers are faster and more powerful — but need to be more energy-efficient

Supercomputers are faster and more powerful — but need to be more energy-efficient
Sustainability3 min read
  • Supercomputers typically consume anywhere between 1 to 10 megawatts of power on average.
  • Architectural and software innovations are critical in driving energy-aware HPC.
Supercomputers have been fuelling some real-world business solutions over the decades. Right from designing a car or an aeroplane to oil field exploration to genome mapping, supercomputers have solved the most complex challenges for humanity.

In recent years, High Performance Computing has seen increased mainstream adoption. It has literally moved out of research labs where only the privileged set of users experimented with the technology. Enterprises of all sizes today leverage HPC to improve time to market and ensure a competitive edge. HPC market is expected to grow from USD 36.0 billion in 2022 to USD 49.9 billion by 2027, at a CAGR of 6.7%, according to MarketsandMarkets.

Supercomputers have certainly become more powerful and faster in the process. With this impressive performance improvement came a giant spike in power consumption. Typically, supercomputers or HPC workloads consume huge amounts of electricity, generate significant heat that affects the lifetime of components and leads to millions of dollars in cooling expenses. A typical supercomputer consumes anywhere between 1 to 10 megawatts of power on average, which is equal to the electricity needs of almost 10,000 homes (a 2019 estimate).

Rising energy consumption is pointed out as the biggest challenge faced by next-gen supercomputers.

“The amount of energy needed in data centres to support current and future computing needs for high performance and machine learning is rising. At the same time, because sophisticated nodes have such a large number of processing units, reliable and secure computing has become a challenge” says Dr. Venkat Mattela, Founder & CEO of Ceremorphic, a semiconductor player in the area of Energy Efficient AI Supercomputing.

Ceremorphic has developed a new architecture for next-generation applications such as AI model training, high-performance computing, drug discovery, and metaverse processing. Its silicon geometry-based architecture, claims the company, satisfies high-performance computing needs in terms of power consumption at scale.

Companies have been focusing on applying new technologies, in computing hardware, architecture and cooling technologies, to achieve ambitious energy efficiency goals in HPC systems. Players believe that the importance of addressing the energy requirements to further accelerate advanced AI applications in the industry.

AMD, for instance, aims to deliver a 30-fold increase in energy efficiency across its high-performance compute (HPC) platforms by 2025. “These applications are essential to scientific research in climate predictions, genomics, and drug discovery, as well as training, AI neural networks for speech recognition, language translation, and expert recommendation systems. The computing demands for these applications are growing exponentially. Fortunately, we believe it is possible to optimize energy use for these and other applications of accelerated compute nodes through architectural innovation,” wrote Mark D. Papermaster, CTO of AMD.

Nevertheless, getting to this ambitious goal will continue to be an uphill task for players. Energy-efficient computing is a multi-dimensional debate in the Exascale computing arena.

“Significant modifications to a supercomputing system are both disruptive and costly since it is an ecosystem. All components of the ecosystem, as well as their linkages, must be considered while building future generations of supercomputers,” says Mattela.

He believes that AI/ML will drive the demand for supercomputing in this decade and going forward. The advent of new machine learning models and the performance demands on big models will push the demands of HPC. “We will be seeing an alternate computing paradigm with optical and with device architectures beyond CMOS (complementary metal-oxide semiconductor) technology” he adds. Nevertheless, much work and research are needed to drive energy-aware HPC forward.

SEE ALSO:

Commercial 5G network is going to transform businesses by improving operations, customer experience, employee efficiency and much more

Three fourths of human population will be covered by data regulations

Digital twins market to surge as e-tailers enhance their offerings

READ MORE ARTICLES ON


Advertisement

Advertisement