China's Lingsheng Supercomputer: Homegrown CPUs Power Quest for World's Fastest

China's Bold Leap: Lingsheng Supercomputer Ushers in Exascale Era with Homegrown Innovation

China has made headlines with the unveiling of the Lingsheng supercomputer, a groundbreaking system poised to claim the title of the world's fastest. Developed entirely with domestic technology, this CPU-only behemoth represents a pinnacle of self-reliance in high-performance computing. Announced by the National Supercomputing Center in Shenzhen (NSCC Shenzhen), Lingsheng promises sustained performance exceeding 2 exaFLOPS—surpassing the current leader, America's El Capitan at 1.8 exaFLOPS. For Chinese universities and research institutions, this development signals a new chapter in computational power, enabling unprecedented simulations in fields like artificial intelligence, materials science, and climate modeling.

The announcement came during a conference in Shenzhen on April 24, 2026, where Lu Yutong, director of NSCC Shenzhen and chief designer, highlighted the system's full-stack independence from foreign hardware. This achievement comes amid ongoing U.S. export restrictions on advanced chips, pushing China to accelerate its domestic semiconductor ecosystem. Lingsheng's design not only dodges these barriers but also sets a benchmark for efficient, scalable computing without relying on graphics processing units (GPUs), which dominate Western exascale systems.

The Architecture Behind Lingsheng's Power

At its core, Lingsheng employs 47,000 high-performance domestic central processing units (CPUs) distributed across 92 compute cabinets. A pilot phase utilized 100 Huawei Kunpeng servers equipped with Arm-based Taishan cores, totaling 12,800 cores, demonstrating early viability. The full configuration scales dramatically, incorporating 20,480 computing nodes, each with dual Armv9-based LX2 processors featuring 304 cores per processor, high-bandwidth memory (HBM), and dedicated data movement engines.

Key innovations include 3D floating-orthogonal computing architecture for optimized parallelism and the world's largest centralized liquid cooling system across 67 storage cabinets. Networking leverages the LingQi high-speed interconnect with dual-plane fat-tree topology, delivering 1.6 Tb/s per node and 10 TB/s aggregate bandwidth. Storage reaches exascale levels at 650 PB over 428 nodes. This CPU-centric approach prioritizes balanced performance for both traditional high-performance computing (HPC) and AI workloads, contrasting with GPU-heavy designs elsewhere.

From National Center to Academic Powerhouse: University Ties

NSCC Shenzhen, the driving force behind Lingsheng, maintains deep connections with higher education. Director Lu Yutong is a professor at Sun Yat-sen University's School of Computer Science and Engineering, bridging supercomputing expertise with academic research. The center collaborates with institutions like Southern University of Science and Technology (SUSTech) in Shenzhen, a hub for innovation where supercomputing supports cutting-edge university projects.

Chinese universities have long relied on national supercomputing centers for resource-intensive research. Lingsheng will provide open access for education and scientific computing in southern China, empowering students and faculty at SYSU, SUSTech, and Tsinghua University affiliates. This integration fosters hands-on training in exascale computing, vital for China's goal of producing world-class researchers.

Domestic CPUs: Phytium Feiteng and Huawei Kunpeng at the Helm

Lingsheng's reliance on homegrown CPUs underscores China's semiconductor progress. While specifics remain guarded, the pilot's Huawei Kunpeng (Arm Taishan) points to scalable Arm architectures. Phytium's Feiteng series, with over 10 million units shipped for national projects, likely plays a role, offering high-core counts optimized for HPC. These processors support FP64 precision essential for scientific simulations, rivaling Intel/AMD in domestic ecosystems despite sanctions.

This self-sufficiency stems from U.S. blacklists targeting entities like Phytium since 2021, compelling innovation. Universities contribute through chip design programs at Tsinghua and Peking University, training the next generation of CPU architects.

Photo by Trnava University on Unsplash

Performance Claims and Global Context

With sustained 2+ ExaFLOPS, Lingsheng aims to reclaim China's supercomputing crown, last held prominently with Sunway TaihuLight in 2016. Unlike El Capitan's AMD GPU-APU hybrid, Lingsheng's CPU focus excels in memory-bound tasks. However, absence from TOP500—China halted submissions in 2019 amid sanctions—means claims await independent verification.

For academia, this positions Chinese universities ahead in HPC benchmarks, accelerating theses in quantum chemistry and astrophysics previously throttled by compute limits.

Transforming University Research: Key Applications

Lingsheng targets nine domains: remote sensing, materials science, bioinformatics, meteorology, fluid dynamics, molecular simulations, life sciences, and AI model training. A recent paper, “Breaking the Training Barrier of Billion-Parameter Universal Machine Learning Interatomic Potentials,” showcases its prowess in materials discovery. Universities like SYSU use such systems for drug design, where exascale power simulates protein folding in hours versus weeks.

In climate modeling, NSCC resources aid Peking University projections for Yangtze floods. Bioinformatics at Fudan University benefits from genomic sequencing at petascale speeds, advancing personalized medicine research.

Challenges: Skepticism and Verification Hurdles

Skeptics note unverified benchmarks and limited domestic CPU competitiveness. x86 alternatives like Hygon lag leaders, and Arm pilots scale uncertainly to exascale. Liquid cooling and interconnects demand flawless integration. Yet, China's track record—Sunway's 93 PetaFLOPS in 2016—suggests feasibility.

Universities mitigate risks via hybrid simulations, preparing students for real-world HPC variability.

Empowering China's Higher Education Landscape

Lingsheng bolsters China's 3,000+ universities, where HPC access was bottlenecked. SUSTech and SYSU gain edge in global rankings, attracting talent. Programs like ASC Student Supercomputer Challenge integrate exascale training. This aligns with 15th Five-Year Plan's focus on elite universities and AI.

Student projects in earthquake modeling or fusion energy now feasible, fostering innovation hubs in Shenzhen's university corridor.

Photo by Datingscout on Unsplash

Global Academic Competition and Future Outlook

As U.S. leads TOP500, Lingsheng challenges parity, spurring international collaborations despite tensions. Chinese universities eye hybrid quantum-HPC, with Tsinghua prototyping. By 2030, expect multiple exascale systems, democratizing compute for 60 million students.

For global scholars, Lingsheng exemplifies resilience, inviting cross-border research via open platforms.

Stakeholder Perspectives and Actionable Insights

Lu Yutong emphasizes controllability: "Fully independently controllable hardware and software stack." Huang Xiaohui affirms surpassing El Capitan. Academics at SYSU predict 10x research output boost.

• Step-by-step for researchers: Apply via NSCC portals for allocations.
• Benefits: Faster publications, grant success.
• Risks: Access queues during peak.

Prospective faculty/students: Target HPC programs at SYSU/SUSTech for Lingsheng access.