OceanLight, the Chinese language exascale supercomputer that has defied itemizing within the Top500, has as soon as once more surged to prominence as a finalist within the Gordon Bell Prize. A yearly award in recognition for “excellent achievement in high-performance computing,” the Gordon bell Prize awards a $10,000 fund to the winner, picked from a gaggle of 5 finalists. OceanLight, operated by the College of Science and Know-how of China, as soon as once more races neck and neck with crowning jewels of Western supercomputing achievement equivalent to Frontier and Summit, in addition to Japan’s Arm-based Fugaku.
OceanLight’s supercomputing chops are spectacular, particularly if you take note of its expertise. In addition to being an exascale-class supercomputer, certainly one of only some on the planet proper now, OceanLight does not have the identical facilities as these created outdoors of closely sanctioned China. For one, China is barred from accessing cutting-edge manufacturing tech. This makes it wholly depending on its inner manufacturing capabilities to supply many of the required elements and expertise that go into constructing such an enormous machine.
Because of this OceanLight has been constructed with applied sciences which have principally been deprecated in comparison with that of the present Greatest GPUs and Greatest CPUs fanatics can get their arms on, not to mention the cutting-edge manufacturing processes and applied sciences powering the likes of Frontier, which already employs AMD-made MI300 APUs. Semiconductor Manufacturing Worldwide Company (SMIC), China’s homegrown CPU producer, constructed OceanLight’s exascale capabilities by leveraging a 14nm course of for its SW26010-Professional CPUs.
In accordance with some very educated math from Nextplatform, every SW26010-Professional CPU can ship 14.03 teraflops of compute at both FP64 or FP32 precision and 55.3 teraflops at BF16 or FP16 precision. Extracting knowledge from the earlier Gordon Bell Prize entry the system took half in, it is estimated that OceanLight incorporates a 107,520 node structure with one SW26010-Professional per node, for a complete of 41.93 million cores.
As every cupboard can help 40 nodes, that works out to 105 cupboards in complete. And in keeping with this yr’s Gordon Bell entry, the supercomputing workload that simulated as many as 2.5 billion atoms and their interactions referred to as as many as 28.1 million of OceanLight’s cores.
China’s leveraging no matter expertise it might probably to advance each its quantum and supercomputing capabilities, and the nation is not shy of the elevated energy consumption (principally coal-based) and supplies requirement that comes from scaling older, much less superior expertise. As soon as once more, the place there’s will and state-backed pockets, there is a approach to plow ahead
