On July 23, Alibaba Cloud officially announced that the Alibaba Zhenwu M890 super node has successfully been adapted to the latest flagship large model Qwen3.8, and it is now available on the Alibaba Cloud BaiLian platform to provide model inference services. This marks the Zhenwu M890 as the first super node system in China to successfully support the stable operation of a large model with over 2 trillion parameters, achieving an important breakthrough in domestic software and hardware collaboration in large model computing scheduling.

As Alibaba's latest ultra-large-scale flagship model, Qwen3.8 has a parameter scale of up to 2.4 trillion. Due to the limitations of traditional AI clusters in terms of inter-node communication bandwidth, there are common performance bottlenecks when dealing with high-throughput, low-latency, and high-concurrency inference demands. To achieve stable and efficient operation of a model at this scale, it usually requires splitting the model parameters and distributing them across dozens or even hundreds of high-speed connected GPU cards.

To address the distributed computing challenges of trillion-parameter models, the Alibaba Zhenwu M890 super node has optimized cluster high-speed interconnection technology and communication architecture, effectively overcoming the transmission limitations of traditional AI clusters, ensuring that the large model runs "stably and quickly" during inference. With this service now launched on the Alibaba Cloud BaiLian platform, enterprises and developers can directly access the efficient trillion-parameter inference capabilities, laying the computing foundation for the commercial deployment of next-generation ultra-large-scale AI applications.