By: Michael Feldman
The Xinhua News Agency has reported that China has launched the prototype of Shuguang, an exascale supercomputer being developed by Dawning Information Industry, also known as Sugon.
According to the Xinhua report, the Shuguang machine is “expected to be put into operation in national supercomputing centers in Shanghai and Shenzhen.” The exascale prototype represents the third such systems put into operation over the last year. As we reported in August, the first two protypes were announced during this past summer.
The first of these, unveiled by National Supercomputing Center in Tianjin, is the forerunner to the Tianhe-3 exascale machine. It’s presumed to be an Arm-based supercomputer, based on Phytium’s Xiaomi platform. The second prototype is the precursor to the Sunway exascale system, which is set to be installed at the National Supercomputing Center in Jinan. The Sunway machine is expected to be based on a future version of the ShenWei processor, the latest version of which is used to power the 93-petaflop Sunway Taihulight supercomputer. However, no details were revealed about the nature of the processors powering the two prototypes.
The same is true for the third prototype announced this week, although the final Shuguang exascale machine is expected to rely on domestically-produced x86 processors. And now that Chinese chipmaker Hygon has such technology in the form of a Zen server CPU license, the most likely path to indigenous x86-based exascale machinery looks like it will be through AMD’s intellectual property. Hygon is already shipping their first locally-made Zen chips into the domestic market, under the name of “Dhyana,” and these could certainly be the basis for the Shuguang prototype.
Hygon’s EPYC license is restricted to the AMD’s first-generation Zen architecture, which is unlikely to be powerful enough to be the foundation of an exascale supercomputer. That suggests the Chinese are already looking ahead to future Zen designs to fuel their x86 needs. AMD’s Zen 2 EPYC chip, codenamed Rome, is already in the pipeline and is scheduled to go into production next year. But for exascale systems, the Zen 3 EPYC platform would be the most logical choice. That chip is on schedule for a 2020 release, which is just in time for China’s initial crop of exascale supercomputers.
Of course, Intel could also offer the Chinese licenses of its x86 technology and considering the escalating nature of the US-China trade war, it might behoove all chipmakers to strike local licensing deals to avoid running afoul of future retaliatory tariffs. On the other hand, the US government could decide to cut off Chinese licensing of advanced US technology based on national security concerns. That would effectively scuttle China’s plans to produce domestically produced x86-based supercomputers, not to mention more mainstream servers. But that’s getting pretty far ahead of ourselves.
The bottom line is that China has managed to launch all the prototypes for its three-pronged exascale strategy and do so at least two years prior to the deployment of the final systems. Whether China meets its 2020 target for at least one of these exascale machines remains to be seen, but having three options upon which to draw would seem to improve its chances.