MUSA SDK: China Responds to CUDA with Its Own Domestic AI Platform | Gpu vs cpu architecture | Hardware store | GPU hardware | Turtles AI

Moore Threads introduces MUSA SDK 4.0.1, a Chinese alternative to NVIDIA’s CUDA, compatible with local and international CPUs, designed for AI and easy code porting.

Key points:

• MUSA SDK 4.0.1 supports Intel, ARM and Chinese CPUs, expanding hardware compatibility.
• MUSIFY makes it easy to migrate CUDA code to the MUSA environment, simplifying adoption for developers.
• Libraries such as muBLAS, muFFT and muThrust provide tools for mathematical operations and parallel processing.
• Support for PyTorch via torch_musa enables the execution of AI models on Moore Threads GPUs.

Moore Threads, a Chinese company founded in 2020 by Zhang Jianzhong, former global vice president of NVIDIA, recently released version 4.0.1 of its MUSA SDK, a domestic alternative to NVIDIA’s CUDA stack. This update expands support to Intel, ARM, and Chinese CPUs such as Hygon, Kylin, and Loongson, facilitating adoption across architectures. The MUSIFY tool enables automatic migration of CUDA code to the MUSA environment, while libraries such as muBLAS, muFFT, and muThrust provide capabilities for mathematical operations and parallel processing. Support for PyTorch via torch_musa enables the execution of AI models on Moore Threads GPUs, highlighting the company’s commitment to creating a complete ecosystem for AI.

Moore Threads aims to be a domestic alternative in the field of GPUs and AI, offering tools and libraries to facilitate the development and migration of existing applications.