China creates an alternative to GPU Ecosystem

Moore Threads introduced an updated version of the Musa SDK software complex, designed to work with graphic processors. The development has become part of the Chinese strategy to create independent technologies against the background of limitations introduced by the United States for the range of components with a high -tech, including AI chips, equipment for the production of microcircuits and HBM memory.

Musa SDK is positioned both by the Nvidia Cuda alternative and supports both the MTT GPU series and Intel processors in combination with Ubuntu and Kylin OS. The platform contains a compiler, a set of libraries for machine learning and mathematical operations, as well as error investigation aids. Special attention is paid to compatibility with existing projects – the Musify tool adjusts the code written for Cuda to Musa Architecture.

Demonstration projects are available on the developer’s website, such as generating images with neural networks, 3D modeling of people and audio processing. Some of the examples work on the Server GPU MTT S3000, which emphasizes the orientation of the platform, not only on research tasks, but also on industrial use.

Analysts note that Musa is still inferior to Cuda in terms of optimization and support, but its development can meet the needs of the local market, in particular in the segment of a small and medium -sized scale. In the long term, the success of the ecosystem will depend on the growth of the Chinese GPU performance and the expansion of the SDK functionality.

Earlier, Cuda alternatives, such as AMD ROCM, also experienced problems due to limited support support. Moore Threads is based on a closed ecosystem, which on the one hand accelerates integration with domestic decisions, but on the other hand reduces the audience outside of China.