The Sunway TaihuLight supercomputer in Wuxi, China
For several years, some in the HPC community have suspected China of sandbagging the world on its true supercomputing capabilities. Those suspicions may have been confirmed with the publication of a research paper last week in which Chinese university researchers reported that 40 million heterogeneous cores within China’s Sunway supercomputer have been directed at a deep learning-based simulation of “quantum many-body problems,” defined as “….pertaining to the properties of microscopic systems made of many interacting particles.”
Which is to say problems of extreme complexity and size.
The reaction of Steve Conway, Senior Advisor, HPC Market Dynamics at Hyperion Research, reflects a shared sentiment among many HPC industry watchers.
“This development adds credibility to the notion that China may have deliberately not reported exascale Linpack results for the November 2021 Top500 list in order to avoid attracting more U.S. Government restrictions,” Conway told us in an email.
The Sunway TaihuLight system was developed by China’s National Research Center of Parallel Computer Engineering & Technology (NRCPC) and installed at the National Supercomputing Center in Wuxi, in China’s Jiangsu province.
According to the most recent TOP500 list of the world’s most powerful supercomputers, Sunway TaihuLight is ranked no. 4 with reported performance of 93 pFLOPS. However, China has reported that level of throughput since at least 2016, when it was ranked the no. 1 system in the world.
The performance required for Sunway’s deep learning-based quantum simulation was not reported.
“We report that a deep learning-based simulation protocol can achieve the solution with state-of-the-art precision in the Hilbert space as large as 21296 for spin system and3144 for fermion system, using an HPC-AI hybrid framework on the new Sunway supercomputer,” stated the researchers, who are from a range of Chinese universities. “With highly scalability up to 40 million heterogeneous (sw26010pro) cores, our applications have measured 94 percent weak scaling efficiency and 72 percent strong scaling efficiency. The accomplishment of this work opens the door to simulate spin models and Fermion models on unprecedented lattice size with extreme high precision.”
While many assume China is under-reporting supercomputing performance, another theory is that China is trailing the U.S. in the exascale race and has not shared recent system performance to avoid embarrassment. Depending on which interpretation of Sunway’s quantum many-body simulation runs you believe, both theories could be supported.
