China unveils GPU-free LineShine supercomputer with 2.45 million domestic CPU cores

What was announced

The National Supercomputing Center in Shenzhen (国家超级计算深圳中心) has deployed a new machine called LineShine, a GPU‑free supercomputer built from 2.45 million domestic CPU cores. It has been reported that the system delivers peak performance of up to 1.54 exaFLOPS and is powered by Armv9‑based LX2 processors developed in China. The project marks an unusual architecture choice: exascale class performance without the accelerators — no discrete GPUs — that dominate modern Western supercomputers.

Why it matters

Why build an exascale machine without GPUs? For Beijing the answer is strategic: independence. US export controls and restrictions on high‑end GPU sales to China have complicated Beijing’s access to accelerator hardware used widely for AI and HPC workloads. Reportedly, Chinese centres are increasingly leaning on domestic CPU designs, custom interconnects and software stacks to close the gap. LineShine is being presented as a milestone in that push — proof that an all‑domestic silicon and system stack can scale to exascale-class throughput for many scientific and engineering workloads.

Bigger picture

The announcement will feed broader questions about competition in advanced computing. Can CPU‑centric systems rival GPU‑accelerated clusters for AI model training? Not always — GPUs remain favored for dense linear algebra and many machine‑learning tasks — but CPUs can excel in traditional simulation and I/O‑bound workloads, and they sidestep geopolitical supply risks. It has been reported that LineShine will be used for national research priorities; its deployment underscores China’s dual track: chasing both homegrown hardware and software ecosystems while adapting to tighter technology exports from the West.