China unveils new HPC software platform

June 30, 2026(MLN): China has introduced a new software platform designed to reduce its dependence on foreign computing ecosystems and strengthen the country's self-reliance in high-performance computing (HPC), marking a significant step toward building a fully indigenous scientific computing environment.

The platform, named Yisuanfangzhou, was unveiled in Beijing on Monday. It was jointly developed by the Computer Network Information Center (CNIC) of the Chinese Academy of Sciences (CAS), the University of Science and Technology of China (USTC), the CAS Institute of Mechanics, and Chinese supercomputing company Sugon, according to China Daily.

Researchers said the platform addresses one of the biggest obstacles in China's HPC ambitions the heavy reliance of scientific software on NVIDIA's CUDA programming ecosystem, which has become the global standard for GPU-accelerated computing over nearly two decades, according to APP.

Speaking at the launch, Wang Yangang, a researcher at CNIC and the project's lead developer, said open-source software forms the foundation of today's digital infrastructure, powering everything from operating systems to advanced scientific applications.

However, he noted that the dominance of CUDA has made it difficult for researchers to migrate existing scientific applications to domestically developed processors.

While China has made considerable progress in developing its own high-performance CPUs and GPUs, Wang said software compatibility remains a major bottleneck, limiting the ability to fully utilize domestic computing hardware.

To overcome this challenge, Yisuanfangzhou provides a comprehensive software toolkit that enables scientific applications to be transferred to Chinese-developed chips with minimal manual effort.

The platform supports algorithm optimization, automated code migration, software adaptation and deployment across domestic computing systems.

One of its key components, Jiuyanshu, is a library of 16 optimized scientific computing algorithms covering areas such as linear algebra, parallel computing, computational fluid dynamics, biological computing and deep learning.

According to the development team, several of these optimized algorithms deliver performance improvements exceeding tenfold compared with conventional implementations on domestic hardware.

The platform also incorporates BoundX, an artificial intelligence-powered code conversion engine that significantly reduces the time required to migrate legacy CUDA-based applications. Developers said tasks that previously required around 10 hours of manual coding can now be completed in approximately 30 minutes, with converted applications ready to run on China's domestic supercomputing infrastructure.

According to the developers, the automated conversion engine achieves a success rate of 71%, outperforming existing domestic migration tools by more than 20 percentage points.

Another major feature is Agent-HiReFlow, an intelligent simulation assistant that allows researchers to create engineering simulations using natural language instead of writing complex code.

The system automatically converts user instructions into physical models, configures simulation parameters, executes calculations, monitors runtime performance and produces visualization-ready outputs.

During internal testing, the platform successfully completed a 10-million-cell hypersonic flow simulation in roughly one hour, demonstrating performance comparable to systems powered by high-end NVIDIA hardware.

The developers said Yisuanfangzhou will be integrated into China's national computing infrastructure, including the Orient Supercomputing System and the National Supercomputing Internet, enabling approved researchers to access domestic computing resources without extensively rewriting existing software.

Looking ahead, the project team plans to expand support for additional Chinese chip architectures, strengthen collaboration with global open-source communities and broaden the platform's applications to emerging fields such as digital twins and AI-driven scientific research.

Industry representatives said developing a robust software ecosystem now represents the next critical phase of China's computing strategy.

Nie Hua, chairman of Sugon's subsidiary Zhongke Controllable Information Industry Co., said the country has largely overcome hardware constraints and must now focus on integrating domestic chips into practical scientific and industrial applications through coordinated software and hardware development.

He added that achieving seamless software-hardware integration would be essential for establishing a globally competitive, self-sufficient computing platform capable of supporting both scientific innovation and national strategic priorities.

Copyright Mettis Link News