High‑Performance Computing: Principles, Evolution, Applications, and Market Landscape

High‑Performance Computing (HPC), also known as supercomputing, refers to powerful computer systems that combine many processing nodes via high‑bandwidth interconnects to solve complex scientific and engineering problems, serving as a strategic national asset.

HPC systems operate using two main processing models: serial processing performed by CPUs and parallel processing that leverages multiple CPUs and GPUs, where GPUs excel at simultaneous arithmetic operations and are widely used in machine‑learning workloads.

Performance is measured in FLOPS (floating‑point operations per second). As of early 2019, top‑tier supercomputers achieved 1.43 × 10 17 FLOPS (peta‑FLOPS), while the next milestone, exa‑FLOPS (10 18 ), would require roughly five million desktop machines each delivering 200 GFLOPS.

Key terminology includes HPC (large‑scale compute clusters), supercomputer (the most advanced HPC system), heterogeneous computing (CPU + GPU), memory, interconnect, peta‑FLOPS, and exa‑FLOPS levels.

HPC enables breakthroughs such as rapid drug discovery, molecular dynamics for new materials, and advanced weather forecasting, and it underpins the fourth industrial revolution by supporting AI, big‑data analytics, and scientific simulations.

The architecture of modern HPC systems combines high‑speed interconnects, parallel file systems, and storage arrays; merely adding more CPUs does not guarantee performance without balanced system design.

Application domains span two categories: (1) scientific simulations like aerospace design, nuclear modeling, and astrophysics; (2) data‑intensive tasks such as AI training, big‑data analysis, finance, genomics, and rendering.

China’s HPC market is expanding rapidly. The 14th Five‑Year Plan and new‑infrastructure initiatives have spurred the construction of national HPC centers, with projected market size exceeding 400 billion CNY in 2022 and a CAGR of about 13 % through 2025.

Major domestic vendors—Lenovo, Sugon (曙光), and Inspur (浪潮)—dominate installation share, with Sugon holding a near‑40 % market share in TOP100 installations by 2019. These companies possess end‑to‑end domestic IP from chips to software.

Policy programs such as the “East‑Data‑West‑Compute” initiative aim to distribute HPC resources across regions, leveraging the West’s abundant renewable energy and cooler climate for efficient data‑center operation.

Definition and importance of HPC

Serial vs. parallel processing architectures

Performance metrics (FLOPS, peta‑ and exa‑FLOPS)

Historical evolution: Cray era and multi‑computer era

Key application scenarios

China’s market growth, policy support, and major vendors