DGCity
— A Window to China

A Chinese Supercomputer Wins the Gordon Bell Prize

#China's Chip Buildout#Supercomputing in China

The ACM Gordon Bell Prize, the highest honour for an application of supercomputing, went to a Chinese team at SC26 in the United States on the afternoon of 19 November local time. The winning calculation ran on Sugon 8000, known in Chinese as Dengfeng, an all-domestic AI supercluster that can scale to 100,000 accelerators.

The work was carried out by the University of Science and Technology of China with several institutes of the Chinese Academy of Sciences and the National High Performance Computing Engineering Research Center. It targeted the iron-molybdenum cofactor of nitrogenase, the enzyme that converts atmospheric nitrogen into ammonia, a strongly correlated electron system that has resisted precise treatment for decades.

The team’s route was to represent the electronic quantum state with a neural network, an artificial-intelligence method, and to extend that approach to a larger strongly correlated system than had been handled before, combining it with massively parallel algorithms. The result reported by the researchers is the first classical solution of the cofactor’s key active space.

Sugon says its machine was the only platform supporting the project through its full research and development cycle, and that the award-winning run used a slice of about 10,000 accelerators to reach what a CPU-based system of comparable footprint would deliver. The same machine hosted a second entry, SD3, a global weather and climate simulation at 1.5-kilometre resolution, which was a finalist for the Gordon Bell Prize for climate modelling.

Prize committees award the Gordon Bell for the science enabled, not for the hardware as such, and that distinction is why the result is being emphasised in China. The country’s supercomputing programme has been criticised as strong on machine building and thinner on published applications. A domestic system carrying an award-winning AI-for-science calculation is the kind of evidence that argument needs, and a second finalist across a different discipline strengthens it.

Some caution is due. The account above rests on Sugon’s announcement and Chinese state media reporting; the ACM citation and the accompanying paper will carry the independent technical detail. The larger question is routine access: whether thousands of research groups can use such machines without a dedicated engineering team, which is what turns a prize into capacity.

Like(0)
未经允许不得转载:Dgcity » A Chinese Supercomputer Wins the Gordon Bell Prize

Sign In

Forgot Password

Sign Up