What’s Next In Supercomputing

AsianScientist (Oct. 04, 2024) – Firmly in the midst of the digital revolution, it is clear that emerging technologies are not just participants but drivers of change—reshaping our understanding of what is possible. Artificial intelligence (AI) is changing how we tackle global challenges like climate change, health and food security, processing colossal data sets at speeds traditional computing can no longer match.

Supercomputing is an integral part of this dynamic landscape, offering the speed, memory and power needed for these specialized tasks. Leading the charge are visionaries like Mark Stickells, Executive Director of Australia’s Pawsey Supercomputing Research Centre; Dr. Piyawut Srichaikul, Director of NSTDA Supercomputer Center (ThaiSC), Thailand; and Professor Satoshi Matsuoka, Director of Japan’s RIKEN Center for Computational Science.

Through their work, they aim to establish Asia-Pacific as a supercomputing hotspot and share insights on what is next in the field, both locally and worldwide.

BOOSTING BANDWIDTH

Supercomputers, traditionally tasked with big data problems like weather modeling, quantum simulations or vast astrophysical studies, are continually pushing boundaries. “Since the Industrial Revolution, we’ve been on a steady path of improving machine performance,” noted Matsuoka, indicating a parallel trajectory for supercomputing. Future systems will boast unprecedented processing capacities to meet burgeoning industry demands.

Asia-Pacific nations are highly competitive in the global supercomputing race, frequently ranking among the top performers alongside the US and Europe, which highlights the region’s growing influence and commitment to opening new frontiers in computational science.

China’s Sunway TaihuLight supercomputer, once the world’s fastest, utilizes its custom architecture for various applications, including climate modeling, life sciences and industrial design. Similarly, the Fugaku supercomputer at RIKEN, which Matsuoka helped develop, achieved a milestone by reaching 442 petaFLOPS in 2020, surpassing other leading supercomputers in AI capabilities by about tenfold.

The foremost challenge now, said Matsuoka, is enhancing data movement. “Supercomputers need to efficiently transfer vast amounts of data, focusing on improving bandwidth, memory and interconnect speeds,” he explained. Many modern AI applications encounter bottlenecks primarily due to data transfer constraints rather than computational power.

Hardware innovations across graphics processing units (GPUs), photonics and wafer-scale architectures can help ease this strain. Matsuoka proposed, “Reducing data travel distances, for instance, by employing techniques like stacking 3D memory devices or utilizing interposers, could lead to significant energy savings.” Moreover, if technical obstacles can be overcome, memory integration directly onto or within the central processing unit (CPU) could streamline data access.

Stickells mentioned that Pawsey integrated cuttingedge GPU accelerators into systems like Zythos and Setonix, marking a shift from their traditional, CPU-centric high-performance computing (HPC) approaches. These upgrades dramatically increase computational capacity while maintaining energy efficiency. Setonix—ranked 15th worldwide and providing up to 50 petaFLOPS of power—is 30 times more powerful than Pawsey’s previous systems, Magnus and Galaxy.

Also, ballooning data volumes require novel highefficiency storage and processing approaches. “At Pawsey, we are exploring in-situ data analysis and visualization techniques to quicken data consumption and reduce storage demands,” Stickells explained.

Meanwhile, Srichaikul anticipates exciting advancements on the horizon: “Next-generation supercomputers will likely be shaped by quantum computing, neuromorphic computing and AI-driven optimization algorithms.” Quantum computing tackles complex problems at unprecedented speeds, while neuromorphic computing enhances AI efficiency by mimicking the human brain.

“The fusion of advanced electronic sensing with AI could enable real-time data processing, essential for the evolution of autonomous systems and smart cities,” Srichaikul concluded, adding, “We’ll just have to wait and see.”

CLEAN COMPUTING

Historically, supercomputers were notorious for energyintensive operations: the Cray XT5 “Jaguar” consumed about 7 megawatts—equivalent to the energy usage of thousands of homes. Today, researchers are not just reaching new heights in speed and power; they are equally dedicated to eco-friendly operations.

According to Stickells, Pawsey’s Setonix supercomputer is both more powerful and strikingly greener. Setonix employs a direct liquid cooling system with closed-loop technology to efficiently remove heat without needing energy-intensive air cooling.

Additionally, its unique groundwater cooling system recirculates heat without a net loss of groundwater, saving “approximately seven million liters of water every year that we would otherwise have to evaporate in cooling towers,” Stickells explained. The facility further reduces its carbon footprint with a 120 kW solar array that supports a zeroemission cooling system and boasts dual-skin insulation to maximize energy retention.

Japan is home to some of the world’s greenest supercomputers. RIKEN’s Fugaku, as Matsuoka illustrated, slashes energy costs and emissions through advanced cooling systems and government-regulated carbon-free electricity. Fugaku is about 70 times more powerful than its predecessor, the K computer, and has nearly 50 times greater power efficiency. Matsuoka also stressed the importance of cultivating energy-saving practices in the research community. When RIKEN users maintain their energy consumption below a specified limit, they receive credits that enhance their priority in the job scheduling queue, incentivizing energy-efficient practices in the institute.

Similarly, Pawsey is promoting environmental consciousness in the supercomputing ecosystem. In 2021, the centre published a paper, Energy-based Accounting Model for Heterogeneous Supercomputers, which, according to Stickells, links service units to energy use and aligns tasks with real-time grid data, helping researchers understand the environmental impacts of their projects.

STRENGTH IN SYNERGY

The ethos of “we’re stronger together” deeply resonates with Asian supercomputing leaders, as affirmed by Srichaikul, who also co-chairs the ASEAN-HPC Task Force. Initiatives like the task force allow member states to tackle large-scale technical barriers that surpass the capabilities of any single nation.

For example, spurred by the urgency of the emerging pandemic, the COVID-19 HPC Consortium formed in 2020 brought together HPC resources from Australia, Japan, Singapore and the US to accelerate research efforts. This unprecedented international alliance provided over 600 petaFLOPS of computing power, enabling researchers worldwide to share data and expertise, significantly advancing efforts to combat the pandemic .

“By sharing infrastructure, expertise and data, we can pool resources, reduce redundancy and boost overall efficiency. This is crucial for the success of ASEAN HPC,” Srichaikul asserted, underscoring the importance of fostering a spirit of unity and mutual benefit to maximize the impact of supercomputing resources.

Echoing these thoughts, Matsuoka commented that developing transformative supercomputing technologies demands a hive mind approach and diverse expertise. He has observed a positive shift in global dynamics, from competition to collaboration. “Things are very different now,” he remarked, adding that RIKEN’s international program partners are being engaged earlier and more integrally.

Stickells, speaking from Pawsey in Perth—among the world’s most remote cities—also stressed the significance of connectivity: “We’re highly connected, partnering with local universities and global entities to enhance research and ensure access to the best infrastructure.”

Investing in this network not only sharpens Australia’s competitive edge but also establishes its leadership in global quantum research and sustainability.

WELCOME TO THE REAL WORLD

Experts anticipate real-world impacts of supercomputing to be profound and far-reaching. On a regional level, Srichaikul suggested that supercomputers could refine infrastructure planning, optimize traffic management, protect food security and strengthen disaster response systems.

However, these benefits will only be realized with seamless integration. “We need to design ‘integrationmechanisms’ to incorporate supercomputing into the value chain that address the complex demands of HPC-clouds, sustainability and energy costs,” Srichaikul warned.

Matsuoka envisions the “post-Moore” era—a shift from increasing computer power by making transistors smaller to leveraging new technologies like quantum computing and neuromorphic computing—reshaping industries and daily lives.

Dramatic improvements in computing capabilities make for smarter devices, improved healthcare frameworks and enhanced digital experiences, for instance. AI with real-time data processing and decision-making capabilities will be essential for autonomous systems and smart cities. Connected urban communities will enable smarter management of traffic and energy, creating greener, more efficient living spaces.

“These technologies, combined with improvements in energy efficiency and materials science, will push the boundaries and demands of what supercomputers have to offer,” confirmed Srichaikul.

The proliferation of HPC is gaining momentum, driven by significant cost reductions. Matsuoka indicated that training and using AI is becoming more affordable due to advancements in cloud computing, more efficient algorithms and increased availability of low-cost hardware , making these technologies more accessible to small businesses and individuals. This could spark new innovation opportunities once reserved for large corporations.

Even so, integrating supercomputing into everyday life requires robust ethical frameworks. Stickells stressed the necessity of responsible AI usage in supercomputing, calling for frameworks to ensure that these advancements benefit society while minimizing potential risks.

The Australian Government recognized this, engaging in industry consultations to support the responsible use of AI. In 2024, it held a public consultation, the Safe and Responsible AI in Australia Consultation, advocating for community-first approaches and transparent, collaborative regulatory frameworks.

NURTURING TOMORROW’S TRAILBLAZERS

People remain at the heart of supercomputing’s evolution. At Pawsey, embracing diversity leads to a richer, more innovative and more effective approach to scientific and technological challenges. This philosophy has shaped their recruitment processes to increase gender diversity and foster an inclusive workplace culture. Stickells sees progress in smashing outdated stereotypes, saying, “At least half of the Pawsey team were born outside Australia and more than one quarter identify as female.”

Educational initiatives in the region demonstrate the sector’s commitment to building a skilled workforce. “The EU-ASEAN HPC School proved that it can be done,” Srichaikul remarked, crediting it with laying the foundation for a robust supercomputing talent pool in Southeast Asia, thanks to contributions from international partners. Now known as the ASEAN HPC School, it continues to thrive, hosted by Indonesia in 2023 and 2024, with plans to rotate among ASEAN member countries.

In Thailand, the ThaiSC supports the AI Human Capacity Development initiative, providing computational resources to the Super AI Engineer Program, now in its fourth season, aimed at reskilling and upskilling the AI workforce. These efforts have bolstered Thailand’s competitiveness in the AI and HPC fields, though Srichaikul advocates for greater integration into the broader AI development landscape.

Internationally, collaborative networks like the Alliance of Supercomputing Centres—a knowledge sharing association that includes member hubs from the US and Europe—foster idea exchanges and support among supercomputing centers.

“Even countries with emerging economies can become leaders in specific AI applications by focusing on localized impact and leveraging global partnerships,” said Srichaikul. This ensures the next generation of ASEAN researchers upholds the legacy of supercomputing innovation in the region.

This article was first published in the print version of Supercomputing Asia, July 2024.Click here to subscribe to Asian Scientist Magazine in print. 

Copyright: Asian Scientist Magazine.

Disclaimer: This article does not necessarily reflect the views of AsianScientist or its staff.

 

Read More

Leave a Reply