China's National Supercomputing Center in Shenzhen has announced the LineShine system, an ambitious all-CPU supercomputer designed to reach 2 exaflops of performance. Once fully constructed, this machine aims to be one of the fastest exascale systems in the world.

Unlike most current leaders on the Top500 list, which rely heavily on Nvidia or AMD GPUs for acceleration, LineShine is planned to be a system of pure processor-induced heft.

The Scale of the New All-CPU Supercomputer

The construction of this system will occur in distinct stages to reach its massive performance goals. The initial phase focuses on establishing a foundation with 100 Huawei Kunpeng servers, which will feature 12,800 cores spread across the cluster.

The project's expansion is even more significant. According to recent reports, the second phase of development involves:

  • Adding an extra 47,000 individual CPUs.
  • Distributing these processors across 92 compute cabinets.

Hardware Architecture and Networking Specs

Detailed technical specifications can be found in a pre-print paper titled "Breaking the Training Barrier of Billion-Parameter Universal Machine Machine Learning Interatomic Potentials." This document reveals the massive scale of the hardware required to power such a system.

The total configuration is expected to consist of 20,480 computing nodes. Each node will be equipped with two Armv9-based LX2 processors, featuring impressive internal architecture:

  • Each LX2 CPU contains two compute dies with a total of 304 cores.
  • The processors will run alongside eight 32 GB HBM stacks.
  • Connectivity is handled by the "LingQi high-speed network," utilizing a dual-plane multi-rail fat-tree topology.
  • This network setup offers a massive 1.6 Tb/s bandwidth per node.

Global Standing and Scientific Applications

If these performance figures are accurate, this all-CPU supercomputer will hold the title of the fastest all-CPU-powered machine ever built. While it may not surpass El Capitan—which currently holds the record for the fastest overall system with a 2.74+ exaflop Rpeak—it represents a massive leap in CPU-centric computing.

The LineShine system is being developed to deliver internationally leading performance in several critical fields. Its primary mission includes:

  • Molecular dynamics and fluid simulation.
  • Advanced life sciences research.
  • Large-scale AI model training and promotion.

While the technical jargon can be dense, the sheer scale of this project is undeniable. One can't help but wonder what the eventual Geekbench score would look like.