China's LingSheng Supercomputer Tops TOP500: How Pure CPU Precision Calculation Pushed El Capitan to Second?

 

    On June 23 in Hamburg, Germany, the global supercomputing "summit showdown" — the TOP500 list — was updated. China's supercomputer LingSheng quietly took the top spot, pushing the U.S. machine El Capitan, which had held the lead for over a year, down to second place.
    This marks the first time since 2017's Sunway TaihuLight that a Chinese supercomputer has reclaimed the world #1 position after more than eight years. Many people's first reaction might be "way ahead"? This does indded have something to do with Huawei, but it's not that simple.
    Some ask: Isn't this just AI computing power? Actually, it's quite different. Others wonder: Where are Google and Microsoft? How did a machine this huge even get scored without being shipped to Germany? Today, let's break it all down clearly and objectively — no hype, just the facts.

1. What Exactly Does TOP500 Test?

    The TOP500 list has been around since 1993 and is updated twice a year (June and November). It's basically the supercomputing world's report card. It only has one test: LINPACK — solving an extremely large system of linear equations and measuring how many floating-point operations can be done per second.
    Two key terms to remember:

• Measured performance (Rmax): 

    Not the theoretical peak performance (Rpeak) that manufacturers advertise, but the actual, stable, long-running result. LingSheng achieved 2.198 exaflops (Rmax) against a 2.736 exaflops peak — extracting about 80% of its theoretical power. That's very strong.

• Double precision (FP64): 

    This is the crucial difference from today's AI computing, which we'll explain in detail shortly.

2. What Is This FP64 Test Actually Calculating?
    Here's a simple analogy: simulating airflow around an airplane wing. The computer divides the air into billions of tiny cubes. The pressure, wind speed, and temperature in each cube are influenced by its neighboring cubes — they push and support each other, creating a massive web of interconnected equations.
    You have to solve billions of these equations simultaneously. Once one frame is done, you move to the next, repeating the process tens or hundreds of thousands of times. That's the real nature of the LINPACK benchmark.
Why does it require FP64 (64-bit double precision)? Because it demands extreme accuracy. The more decimal places you keep, the smaller the error. With lower precision, errors accumulate like a snowball rolling downhill. Eventually, the simulated airplane could "fall apart" inside the computer. Scientific simulations need precision down to many decimal places, which is why 64-bit is required.

3.Does the Competition Have Weight Classes?
    Unlike boxing, TOP500 has no weight classes or size limits. The bigger, more power-hungry, and more expensive the machine, the higher the score. It naturally favors well-funded national teams.
    However, there's a sister list called Green500 that measures energy efficiency (how many operations per watt). LingSheng performs noticeably worse than El Capitan in this category.

4.How Big and Power-Hungry Is LingSheng Actually?
    This machine is massive:

• Computing speed: 2.198 exaflops of double-precision performance (currently the world's fastest)
• Processor: LX2 (ARMv9 architecture), 304 cores per chip, two chips per node
• Scale: Over 20,000 nodes, more than 40,000 CPUs, totaling 13.79 million cores, housed in 92 cabinets
• Power consumption: 42 megawatts at full load — about 370 million kWh per year, costing over 200 million RMB in electricity (at Chinese rates), plus dedicated substations and a full liquid cooling system

Classic "brute force wins" engineering.

5. How Are Scores Submitted and Verified?
    The machine is far too big to ship to Germany. All runs happen domestically, results are submitted, and officials conduct spot checks. This voluntary submission system has been in place since 1993.
    There are four layers of anti-cheating protection:

• The test itself includes built-in verification — if the error exceeds 16, the result is automatically invalidated.

• A single global scoring standard.

• Official verification rights with random on-site audits.

• Reputation — the community is small, and getting caught faking data would embarrass the entire institution and country.

    Power consumption is also measured for the Green500 rankings (full system or scaled node tests). In short, it's more like submitting a reproducible, auditable scientific report.

6. What Is the U.S. #2 Machine, El Capitan?
    El Capitan belongs to the Lawrence Livermore National Laboratory and was built by HPE Cray. It has 11.34 million cores and consumes 29.7 megawatts. Its main job is simulating nuclear explosions (real nuclear testing is banned).
    The biggest architectural difference: LingSheng is pure CPU, while El Capitan is a CPU + GPU hybrid (using AMD MI300A APUs). Result: LingSheng's 2.198 exaflops beat El Capitan's 1.809, but it uses about 40% more power for roughly 20% more performance. It wins on raw speed, but trails slightly on efficiency.
    LingSheng's standout feature is that it's fully domestically produced — both the CPU and operating system.

7. What’s the Story Behind LingSheng’s CPU?
    The LX2 CPU uses ARMv9 architecture. Huawei participated in its design and it shares lineage with the Kunpeng series. Each chip has 304 cores, with 32GB HBM on-package plus up to 256GB DDR5 memory.
    China can now produce HBM (companies like CXMT are advancing it). Why skip GPUs? The CPU already includes ARM vector and matrix units, so it can handle part of what GPUs do — "good enough," though not dominant.
    The official foundry hasn't been disclosed. Given current supply chain realities, it's likely produced on a domestic advanced process node, which also helps explain the higher power draw.
    Can regular users buy an LX2 CPU? No — it's a custom supercomputing design. However, Kunpeng-architecture server chips are commercially available, so organizations can build systems on this ecosystem.

8. Is This the Same as AI Computing Power?Not at all.
    Modern AI focuses on speed and volume — using low-precision formats like FP16, FP8, or even FP4 to process more numbers at once while tolerating some error.
    Supercomputing LINPACK demands precision — FP64 double precision with minimal error. Microsoft's Azure machine participated and ranked around #32. Google didn't enter, as this benchmark isn't closely tied to their AI work.
    If mixed low-precision were allowed, El Capitan's GPUs could potentially win. But when limited to pure FP64, LingSheng comes out on top.
    Think of it this way: supercomputers are like Formula 1 cars — optimized for ultimate precision. AI data centers are like massive cargo fleets — built for throughput. Different races, different rules.

9. What Do China's Supercomputers Usually Do?
    LingSheng, from the Shenzhen Supercomputing Center, mainly handles weather forecasting and long-term climate simulation.
    China operates a network of supercomputing centers serving industry and research:

• Tianjin (Tianhe-1 & Tianhe-3): Petroleum, aerospace, biomedicine, controlled nuclear fusion, high-end equipment, animation rendering
• Guangzhou (Tianhe-2): Biomedicine, automotive, shipbuilding, film/animation, finance, nuclear power, oceanography, public safety
• Wuxi (Sunway TaihuLight): Oceanography, oil & gas, climate, industrial design, animation rendering (the 2016 fully domestic champion)
• Jinan (Sunway Blue Light): Oceanography, modern agriculture, oil & gas, drug screening, finance
• Zhengzhou (Songshan): Digital economy, precision medicine, biological breeding, environment, AI-related computing
• Kunshan (Yangtze River Delta Major Science Facility): AI, biomedicine, materials, atmosphere, oceanography

    China's model treats supercomputers more like public infrastructure ("water, electricity, and coal") for science and industry. The U.S. tends to reserve its most powerful machines for classified defense work. Two different philosophies.

10. How DogAI and Supercomputing Actually Work Together?
    Running large AI models directly on supercomputers? Not practical — it would be painfully slow (all CPU, no GPU acceleration).
    The smart combination works in two steps:

• AI writes the code, supercomputer runs it: 

    AI is several times more efficient at coding. Once the program is ready, the supercomputer handles the high-precision FP64 simulations that require extreme accuracy.

• Supercomputer generates training data for AI (the "distillation" approach): 

    The supercomputer first runs high-fidelity simulations (earthquakes, weather, nuclear events, etc.) and produces massive amounts of precise data. This data is then fed to AI models for training. Afterward, the AI can quickly approximate results.
    It's like the supercomputer calculates the full "multiplication table," then the AI "student" memorizes it and can do fast mental math later. Supercomputers provide high-quality "textbooks"; AI provides fast "application."

11. Can NVIDIA Just Take Over This Work?Not quite yet.
    NVIDIA has deliberately de-emphasized FP64 performance in its newer chips to optimize for low-precision, high-throughput AI workloads. The latest GB300 generation actually has weaker FP64 capability than some earlier NVIDIA chips. The design directions are fundamentally different.

12. A Final Note
    LingSheng's #1 ranking carries real weight: fully domestic, pure CPU, ARM architecture, and #1 in FP64 scientific computing.
But don't let the "world #1" headline mislead you. It won on the scientific computing track — that's a completely different race from AI capability. The future isn't "AI or supercomputing." It's complementary:AI handles coding and pattern recognition,Supercomputers run high-precision simulations,Supercomputer results feed back to improve AI training.

    Reclaiming the top spot after eight years reflects tremendous effort and is worth recognizing. At the same time, we should stay rational — China has made important progress in supercomputing, but there's still a long way to go. Keep pushing forward.