'The most powerful machine yet assembled on Earth': China quietly launched record-breaking supercomputer — and someone virtually deconstructed its processor

HPC
(Image credit: Shutterstock / Connect World)

China has quietly launched the Tianhe-3 supercomputer, which is believed to be the most powerful machine currently in existence. 

The machine, built for the National Supercomputer Center in Guangzhou, has been shrouded in secrecy (as you would expect from a supercomputer developed and built in China), sparking plenty of speculation.

The Tianhe-3, also known as "Xingyi," is thought to be a significant leap forward in supercomputing technology, potentially surpassing the capabilities of the upcoming "El Capitan" supercomputer being developed by Hewlett Packard Enterprise and AMD for the Lawrence Livermore National Laboratory. 

Virtually deconstructing Tianhe-3's processor

In November, 2023, TheNextPlatform ran an analysis of the Top500 supercomputer rankings which suggested that the Tianhe-3 could have a peak performance of 2.05 exaflops and a sustained performance of 1.57 exaflops on High Performance LINPACK. This, the site said, would make it the “most powerful machine yet assembled on Earth”. 

The Tianhe-3 is the latest in a series of supercomputers built by the National University of Defense Technology in China. Its predecessors, the Tianhe-1 and Tianhe-2, also made significant impacts on the supercomputing world, with the Tianhe-2 still ranking among the top 30 supercomputers even after several years of operation. 

One of the most intriguing aspects of the Tianhe-3 is its processor. A recent case study on programming the Matrix-3000 (MT-3000) accelerators, submitted to arXiv, provided some insight into the machine's architecture. Delving into this, TheNextPlatform concluded that the Tianhe-3 uses a hybrid device with CPU and accelerator compute as well as three different kinds of memory, two of which are located in the compute complex.

The site said: “It is [more] akin to the AMD “Antares” MI300A CPU-GPU hybrid that is going into El Capitan than it is like the discrete CPU-GPU systems we see pushing the flops in AI and HPC systems these days. The MT-3000 is its own animal, and you might assume that it uses a chiplet packaging architecture given that Semiconductor Manufacturing International Corp (SMIC), the indigenous Chinese foundry, probably could not cram enough transistors into a 14 nanometer process to make a monolithic die. But, then again, maybe this is a 10 nanometer or even a 7 nanometer device. If NUDT doesn’t care about cost, then the yield can be terrible so long as SMIC can find tens of thousands of good MT-3000 parts to make the system.”

The Tianhe-3 supercomputer will support various application scenarios, including high-performance computing, AI large model training, and big data analysis. It is expected to boost the multi-field application service capabilities of the National Supercomputing Guangzhou Center, providing services to Guangzhou City and Guangdong Province.

More from TechRadar Pro

Wayne Williams
Editor

Wayne Williams is a freelancer writing news for TechRadar Pro. He has been writing about computers, technology, and the web for 30 years. In that time he wrote for most of the UK’s PC magazines, and launched, edited and published a number of them too.