Chinese computer researchers have made a major breakthrough by improving supercomputing performance using domestically made graphics processors. Their system achieved a near-tenfold boost compared to Nvidia-powered supercomputers in the United States, according to a peer-reviewed study.
This achievement challenges the dominance of American-made chips and highlights the impact of Washington’s increasing tech sanctions on China’s technological advancements. For years, the United States has been a global leader in supercomputing technology, largely due to powerful processors made by companies like Nvidia.
However, Chinese scientists have now demonstrated that innovative software optimization can unlock greater efficiency from domestically produced chips. The study suggests that by refining software and improving data transfer methods, China’s supercomputers can outperform even some of the best US-based systems in specific scientific computations.
According to reports by scmp, the latest achievement is part of China’s broader strategy to reduce its dependence on Western technology. The country has faced significant challenges due to US sanctions, which have limited its access to advanced semiconductor technology. The development of high-performance computing solutions using domestic hardware is a crucial step in overcoming these obstacles.
Supercomputing is essential for many real-world applications, including weather forecasting, flood control, and urban planning. These fields require massive computational power to run high-resolution simulations, which can take significant time and resources. The reliance on foreign-made processors has been a long-standing issue for Chinese researchers, but this breakthrough may help them move toward greater technological independence.
Professor Nan Tongchao from the State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering at Hohai University in Nanjing led the research team responsible for this innovation. His team developed a new parallel computing approach called the “multi-node, multi-GPU” method, which significantly improves data processing efficiency.
The researchers optimized data transfer and task coordination between multiple computing nodes to minimize performance loss. By making better use of available domestic GPUs and CPUs, their system was able to outperform traditional supercomputers in certain tasks.
In 2021, the Oak Ridge National Laboratory in the US introduced a flood forecasting model called TRITON using the Summit supercomputer. TRITON used 64 computing nodes but achieved only a sixfold speed increase. In contrast, the Chinese model achieved the same speed increase using just seven nodes, reducing node usage by nearly 89 per cent.
The new computing model was built on a domestic general-purpose x86 computing platform. It used Hygon processors with 32 cores, 64 threads, and a 2.5 GHz clock speed. The GPUs were also domestically produced, supported by 128GB of memory and a network bandwidth of 200 Gb/s.
The research team tested their model by simulating flood evolution at the Zhuangli Reservoir in Shandong province. Using 200 computing nodes and 800 GPUs, they completed the simulation in just three minutes—achieving a speedup of over 160 times compared to previous methods.
Professor Nan emphasized the importance of this breakthrough for real-world applications. Faster and more detailed simulations can significantly improve flood control, disaster prevention, and reservoir management. By reducing the time needed for complex simulations, authorities can make better decisions in emergencies, ultimately saving lives and property.
The researchers have made their software open-source, allowing others to build upon their work. Nan believes that this computing model could be useful in other areas, including hydrometeorology, sedimentation studies, and groundwater interactions. The team plans to expand its applications and further test the model’s stability in engineering projects.
This achievement demonstrates that innovation can thrive even under strict technological restrictions. With further advancements, China may soon rival the US in cutting-edge computing technologies. By focusing on homegrown solutions, China is gradually reducing its reliance on foreign technology.
If these trends continue, the balance of power in the tech industry could shift, with China emerging as a major player in the supercomputing sector. The study detailing this breakthrough was published in the Chinese Journal of Hydraulic Engineering on January 3, marking an important step forward in China’s technological journey.
