More than three hundred university teams from around the world competed in the ASC26 Student Supercomputer Challenge Finals in Wuxi, China, where a 5,000-watt power limit required competitors to prioritize energy efficiency alongside performance. Twenty-five finalist teams tackled demanding workloads, including the UnifoLM-WMA-0 world model, AMSS-NCKU gravitational wave numerical simulation, and QiboTN quantum circuit simulation. Peking University won the championship, demonstrating success across challenges in artificial intelligence and quantum computing. The competition uniquely emphasized collaboration; finalist teams were divided into five groups to jointly optimize the Gordon Bell Prize-winning ICON model and recreate a digital twin of Earth.
ASC26 Competition: Workloads and Benchmarks Featured
The competition, culminating in finals held in Wuxi, China, attracted over three hundred university teams globally, narrowing the field to twenty-five finalists who showcased their designs and skills. This broadened scope reflected a deliberate effort to assess capabilities beyond traditional high-performance computing metrics. ASC26 also incorporated a collaborative element, emphasizing teamwork and applying a globally recognized climate model to underscore the competition’s commitment to real-world problem-solving. The Peking University team, winning the championship, demonstrated proficiency across multiple challenges, including world models and quantum circuit simulation, highlighting “the deep understanding of artificial intelligence, supercomputing systems, and interdisciplinary applications” of the next generation of technology talent, according to event organizers.
The e-Prize Challenge showcased innovative approaches to inference optimization; the Peking University team achieved success with UnifoLM-WMA-0 through “extensive low-level code restructuring to build an exceptionally compact, purely functional inference stack.” Their work involved customizing the underlying graph compilation framework and implementing a novel caching strategy to balance speed and quality. Jack Dongarra, Chair of the ASC Advisory Committee, noted that high-performance computing and artificial intelligence are rapidly converging, and that ASC provides an exceptional platform for this growth, challenging students to achieve breakthroughs under constraints and translate ideas into tangible results. Zhang Yonghong, Secretary of the Party Committee at Wuxi University, stated that hosting the ASC26 Finals “represents an important opportunity for the development of Wuxi University.”
Peking University’s UnifoLM-WMA-0 Inference Optimization
High-performance computing increasingly demands optimization beyond raw processing power, with a growing emphasis on energy efficiency and specialized workloads. Recent competitions, like the ASC26 Student Supercomputer Challenge, reflect this shift, moving away from solely benchmarking traditional HPC applications toward more complex tasks such as artificial intelligence and scientific simulations. Teams are now challenged to build powerful systems within strict power constraints, such as the 5,000-watt limit imposed on all clusters at the ASC26 Finals. This constraint forces a re-evaluation of design choices, prioritizing efficiency alongside performance, a departure from conventional supercomputing benchmarks. Peking University’s success in the ASC26 e-Prize Challenge, focused on inference optimization for the UnifoLM-WMA-0 world model, showcased an innovative approach. This granular level of optimization, combined with a novel caching strategy, allowed them to “successfully balance inference speed and output quality,” according to the competition results.
These challenges extend beyond standard benchmarks, probing the capabilities of these mini-supercomputers in areas crucial to contemporary scientific advancement. The team’s approach to UnifoLM-WMA-0 demonstrates a trend of tailoring hardware and software to the specific demands of increasingly sophisticated AI models.
Hosting the ASC26 Finals represents an important opportunity for the development of Wuxi University. In recent years, the university has focused on modern industrial clusters, supercomputing, artificial intelligence, and other frontier fields, while building an integrated talent development system that combines industry, education, and regional collaboration.
Zhang Yonghong, Secretary of the Party Committee, Wuxi University
