The National Supercomputing Center in Shenzhen has introduced LineShine, a high-performance computing system claiming a sustained performance of 2 Exaflops. The machine distinguishes itself by its reliance on an entirely domestic supply chain, utilizing over 2.45 million CPU cores across 20,480 nodes without the inclusion of specialized graphics processing units (GPUs).
LineShine signals a shift in architectural philosophy, prioritizing a CPU-only design to eliminate data bottlenecks common in systems that shuffle information between processors and accelerators.
Technical Composition
The system utilizes a high-speed, dual-plane multi-rail fat-tree network dubbed Lingqu, facilitating 1.6 Tb/s of bandwidth per node. The hardware stack comprises:
Processors: Each node houses two Armv9-based LX2 processors (reportedly associated with Huawei's architecture).
Core Count: Totaling 40,960 processors, reaching more than 2.45 million individual cores.
Memory & Storage: Each processor integrates eight HBM (High Bandwidth Memory) stacks, providing 32 GB of capacity and 4 TB/s of aggregate bandwidth per unit, supported by a 650 PB storage infrastructure.
Strategic Autonomy
The development of LineShine acts as a public demonstration of technological insulation. By excluding foreign-made components, the project aims to render international export controls ineffective regarding its high-end computing ambitions.
"The successful completion of the full-machine testing… demonstrates its complete self-reliance and controllability across the entire stack," stated Li Xiaoli, Deputy Director of the Shenzhen Science and Technology Innovation Bureau.
| Feature | LineShine Specifications |
|---|---|
| Architecture | All-CPU (Armv9 LX2) |
| Performance | ~2 Exaflops |
| Network | Lingqu (Fat-tree topology) |
| Foreign Reliance | Zero |
Background and Context
The project serves as a direct response to global competition in the TOP500 rankings, where machines like the US-based El Capitan—which relies on AMD GPU accelerators—currently dominate the performance spectrum. While standard high-performance computing has leaned heavily on GPU acceleration to hit exascale milestones, LineShine explores whether dense CPU-only scaling can match or exceed these benchmarks.
Read More: RX 9070 XT GPU Price Drops to $629 in Amazon Sale
Beyond raw speed, the system is designed to facilitate a "super-intelligent fusion" platform, intending to consolidate diverse workloads ranging from molecular dynamics and fluid simulations to the training of large-scale artificial intelligence models. Whether this massive parallelization of CPU cores can overcome the inherent efficiency of GPU-accelerated clusters in large-model training remains the primary point of technical uncertainty for industry observers.
