The MUSA parallel computing software now allows for CUDA code porting and supports Intel, ARM, and domestic CPUs such as Hygon and Loongson. This is an ambitious move for China’s technological sovereignty.
The race for technological independence in the fields of artificial intelligence and high-performance computing marks a new milestone with the latest update of the MUSA SDK (Moore Threads Unified Software Architecture). This solution developed by the Chinese company Moore Threads—known for manufacturing domestic GPUs—reinforces its role as a genuine alternative to the NVIDIA CUDA ecosystem, whose dominance in the sector has been virtually unquestioned until now.
The new version MUSA SDK 4.0.1 includes support for Intel and ARM processors, as well as Chinese architectures like Hygon, Loongson, and Kylin, solidifying its compatibility in both international and local environments. However, the significant innovation lies in its MUSIFY portability tool, which allows for migrating CUDA code to MUSA, paving the way for a less disruptive transition for developers looking to move away from the NVIDIA ecosystem.
Technical Comparison: MUSA SDK vs CUDA
Below is a table presenting a realistic comparison between MUSA and CUDA in their key aspects:
| Feature | MUSA SDK (Moore Threads) | CUDA (NVIDIA) |
|---|---|---|
| GPU Compatibility | Only Moore Threads GPUs | NVIDIA GPUs (GeForce, RTX, Tesla, etc.) |
| CPU Compatibility | Intel, ARM, Hygon, Loongson, Kylin | Intel, AMD, ARM (with specific CUDA Toolkit) |
| Main Language | C/C++, Python (limited), support for portability from CUDA | C/C++, Fortran, Python, MATLAB |
| Portability Tool | MUSIFY (CUDA → MUSA) | No reverse portability allowed |
| Mathematical Libraries | muBLAS, muFFT, muThrust | cuBLAS, cuFFT, Thrust, cuDNN |
| Environment | In development, focused on AI, visualization, and HPC in China | Mature, widely adopted in global sectors |
| Updates | Moderate frequency, accelerating with state support | Constant and optimized for new GPU releases |
| License and Access | Open, oriented towards technological sovereignty | Proprietary, tied to the NVIDIA ecosystem |
| Commercial Use | Expanding in universities, public institutions, and Chinese OEMs | Massive in science, industry, automotive, healthcare, AI |
A Strategic Alternative, Not Just Technical
The primary virtue of the MUSA SDK lies not only in its technical architecture or compatibility, but in its geostrategic role. In a context where trade tensions between the U.S. and China have limited access to advanced chips and software, having a platform like MUSA enables the Chinese industry to advance in autonomous and sovereign computing, especially in key sectors such as defense, telecommunications, biotechnology, or energy.
The MUSIFY tool, included in the SDK, offers a direct bridge between developers accustomed to CUDA language and the new Moore Threads environment. This allows for a gradual migration and drastically reduces transition costs. Additionally, the support for essential libraries like muBLAS (equivalent to cuBLAS) or muFFT positions MUSA to provide essential functionalities for scientific and artificial intelligence workloads.
On the Path to a National Ecosystem
Although CUDA continues to be the global de facto standard, the investments by Moore Threads and other Chinese firms in native solutions could shift the balance in the medium term. Some experts believe that the development of tools like MUSA responds to a long-term strategy that prioritizes autonomy over immediate efficiency.
Moore Threads, backed by millions in investments from the Chinese government, aims to create a solid alternative at the hardware and software level, similar to how Huawei is developing its HarmonyOS ecosystem or how Loongson is building its CPUs.
Conclusion
MUSA SDK 4.0.1 represents a strong step towards China’s technological independence in high-performance computing. Although it is still far from the maturity and penetration of CUDA, its rapid evolution, focus on national architectures, and the ease of porting code make it a strategic tool for the future of AI in the Asian country.
The challenge to NVIDIA is no longer just economic or commercial: it is also political and technological. And MUSA intends to play that role as a catalyst for change.
Source: AI News