Jack Dongarra on Supercomputing's AI & Quantum Future

High-performance supercomputing (HPC), once primarily the domain of scientific research, has evolved into a pivotal strategic resource, particularly for training increasingly complex artificial intelligence (AI) models. This convergence of AI and HPC is not only reshaping these technologies but also fundamentally altering how knowledge is generated and its strategic global positioning.

Jack Dongarra, a distinguished US computer scientist and recipient of the 2021 Turing Award for his foundational contributions to HPC software over four decades, recently shared his insights on the future evolution of supercomputing. Speaking at the 74th Nobel Laureate Meeting in Lindau, Germany, Dongarra addressed the transformative roles of AI and quantum computing, along with the broader geopolitical and technological landscapes.

AI’s Pervasive Impact

Dongarra emphasized AI’s already significant role in scientific discovery. He noted that AI is widely used to assist in scientific exploration, particularly for approximating how phenomena behave. “I think of AI as a way to get an approximation, and then maybe refine the approximation with the traditional techniques,” he explained. For demanding problems, where supercomputers traditionally provide solutions through modeling and simulation, AI promises to make these processes “faster, better, more efficient.”

Beyond science, Dongarra believes AI’s impact will be even more profound than that of the internet. He envisions AI becoming incredibly pervasive, serving purposes yet to be fully discovered, and ultimately playing a more substantial role in daily life than the internet has over the past two decades.

Quantum Computing: A Promising Yet Nascent Field

While acknowledging quantum computing as an intriguing area for research, Dongarra cautioned that it has “a long way to go.” He described current quantum hardware as “very primitive” compared to traditional digital computers. Unlike digital systems that yield a definitive answer, quantum computers provide a probability distribution of potential solutions, requiring multiple “runs” to infer an outcome.

Dongarra believes the field has been “oversold,” leading to excessive hype. He predicts a “quantum winter,” a period of disillusionment similar to what AI experienced before its current resurgence. For quantum computing to become truly competitive, significant challenges must be overcome. Quantum computers are highly susceptible to disturbances, leading to frequent “faults” due to the fragility of their computations. Until these systems can be made more fault-tolerant, their practical utility will remain limited. Dongarra expressed skepticism about quantum laptops ever becoming a reality, at least within his lifetime. Furthermore, the development of effective quantum algorithms remains in its infancy, along with the necessary software and infrastructure.

The Future of Computing Architectures

Looking ahead, Dongarra foresees a future where today’s powerful digital supercomputers, often augmented with accelerators like GPUs, will integrate even more diverse technologies. He suggested that quantum computing could become another specialized accelerator, alongside emerging technologies such as neuromorphic computing (which mimics the brain’s structure) and optical computing. Optical computers, which perform calculations at the speed of light by manipulating light beams, offer incredible speed for specific tasks like multiplication. Dongarra envisions a hybridized computing landscape where CPUs, GPUs, quantum devices, neuromorphic processors, and optical components could all combine to tackle complex problems.

Geopolitics and Tech Development

The ongoing geopolitical competition, particularly between the United States and China, is significantly influencing technology development and sharing. Dongarra observed that while the US has imposed restrictions on certain computing technologies, such as Nvidia parts, from being sold to China, unofficial pathways exist, allowing Chinese colleagues access to restricted hardware.

Paradoxically, these restrictions may have inadvertently propelled China’s indigenous technological development. China has shifted focus from acquiring Western technology to heavily investing in its own research and manufacturing capabilities. Dongarra noted that China now designs its own competitive chips and has built powerful supercomputers that likely rival the most significant machines in the US, though information about their performance is not publicly benchmarked. He acknowledged that China’s chip fabrication technology is currently a generation or two behind leading manufacturers like TSMC (Taiwan Semiconductor Manufacturing Company) but expects China to catch up. He also highlighted the complex reality of chip fabrication, noting that some Chinese chips might still be produced in Taiwan, which China considers part of its territory.

AI’s Impact on Programmers

Regarding the future of programming, Dongarra sees AI playing a crucial role in automating time-consuming development tasks. He has been impressed by AI’s ability to generate and optimize software from natural language prompts. He believes that increasingly, developers will describe their desired program functions using everyday language, allowing AI to write the code. While acknowledging potential issues like “hallucinations” or incorrect outputs, Dongarra stressed the importance of building in verification checks to ensure the accuracy of AI-generated solutions. Despite these challenges, he advocates for embracing AI’s potential to transform software development.