Today, Moore Threads announced that it has completed comprehensive adaptation of Alibaba's latest large model Qwen3.5 on its flagship AI training and inference integrated full-featured GPU MTT S5000. This support fully demonstrates the maturity and completeness of the Moore Threads MUSA ecosystem, allowing developers to efficiently complete model deployment and optimization using the MUSA C programming language and Triton-MUSA toolchain.
During the adaptation process of Qwen3.5, Moore Threads validated two core capabilities of the MUSA ecosystem: native MUSA C support allows developers to directly use MUSA C for kernel development, significantly lowering the migration threshold from the CUDA ecosystem; deep compatibility with Triton-MUSA enables developers to write high-performance operators using familiar Triton syntax, which can seamlessly run on Moore Threads' full-featured GPU through the Triton-MUSA backend.
For the hybrid attention mechanism used in the Qwen3.5 multimodal model, Moore Threads has implemented native optimizations. Based on the muDNN computing library and the MATE open-source operator library, it provides efficient support for long sequence processing in the hybrid attention mechanism, successfully achieving high-performance inference of the model on the MTT S5000. This achievement not only reaffirms the extensive adaptability and efficient support capabilities of domestic full-featured GPU computing platforms for cutting-edge large models but also showcases the significant results of software and hardware collaborative optimization.
From GLM-5 to MiniMax M2.5, and now to Qwen3.5, Moore Threads has consistently maintained rapid follow-up and adaptation to top domestic large models. This normalized agile response mechanism stems not only from the seamless compatibility of the MUSA architecture with mainstream AI ecosystems and the continuous optimization of toolchain support but also signifies that the domestic computing foundation has the full-link support capability from model adaptation to efficient deployment. In the future, Moore Threads will continue to deepen its MUSA technology ecosystem, providing a more solid and user-friendly domestic computing foundation to help more cutting-edge large models land applications in a timely manner, accelerating the prosperous construction of the domestic computing ecosystem.
Source: Moore Threads WeChat Official Account

BABA

KBAB

688795

09988

BABX

Moore Threads announced that it has completed comprehensive adaptation of Alibaba's latest large model Qwen3.5 on its flagship AI training and inference all-in-one GPU MTT S5000. This support fully demonstrates the maturity and completeness of the Moore Threads MUSA ecosystem, allowing developers to efficiently complete model deployment and optimization using the MUSA C programming language and Triton-MUSA toolchain

- Today, Moore Threads announced full support for Alibaba's latest model Qwen3.5 on its flagship AI GPU, the MTT S5000.  
- This integration showcases the maturity of the MUSA ecosystem, allowing developers to efficiently deploy and optimize models using MUSA C and the Triton-MUSA toolchain.  
- Moore Threads has consistently adapted to leading domestic models, demonstrating the comprehensive support and optimization capabilities of its domestic GPU platform for advanced models.

Moore Threads quickly completes full adaptation to the Qwen 3.5 model