China activated a national supercomputing internet core node: a single
100,000‑card full‑precision compute pool designed to support concurrent
100,000‑card training, ~10‑trillion‑parameter model full training,
AI‑for‑Science and ultra‑large scientific engineering workloads, and to serve as
a domestically developed inclusive large‑scale compute base. During trial it
completed more than 70 tests at the 10k‑card scale. Using 45,000 cards it ran a
41.47 billion‑atom DFT‑precision simulation, setting a world record;
billion‑scale grid simulation turnaround improved from week‑level to hour‑level;
it has deeply adapted over 400 industry models and completed domestic porting
and optimization of key domain software for fluid dynamics, quantum mechanics
and large‑scale simulation, strengthening the domestic compute application
ecosystem.
