Alibaba Cloud Lingjun Zhenwu M890 supernode instance officially launched

The Zhitong Finance App learned that on August 12, Alibaba Cloud's Lingjun Zhenwu M890 supernode instance was officially launched, and the first batch was sold in the Ulanqab region. Enterprise customers can open a 64-card high-speed interconnect computing power unit on the cloud without building their own computer room, which can carry up to 10 trillion parameter-level MoE large model inference. The Lingjun Zhenwu M890 supernode is also the first supernode type computing power to successfully run a model with over 2 trillion parameters in China. Both Kimi K3 and Qwen3.8-Max have already provided external services through this example.

The Lingjun Zhenwu M890 supernode example supports FP8/FP4 low accuracy calculation. The scale of the scale-up interconnect was raised from 16 karat to 64 cards through the ICN Switch 1.0 chip, and the inter-card interconnection reached 800Gb/s. In training scenarios such as intelligent driving and physical intelligence, the training performance of the previous generation Zhenwu 810E is 3 times better.

The supernode cluster is fully supported by Alibaba Cloud's unified platform, the Lingjun Intelligent Computing Platform. Relying on the HPN 8.0 integrated network, a single cluster supports up to 130,000 cards of heterogeneous computing power mix, which can be expanded to one million cards.

Ulanqab, the region where this supernode example was launched, is one of Alibaba Cloud's top five super data centers. Green electricity accounts for about 90%, which can provide a lower carbon operating environment for high-density computing power. At the same time, the data center pioneered a full-module design architecture. The overall modularization rate for power supply, cooling, security, intelligence and fire protection is as high as 90%, which can greatly shorten the data center delivery cycle, maximize compatibility with mainstream heterogeneous chips and computing architectures, and support at least three generations of chip iterations.

This year, Alibaba Cloud is planning to more than double the global production capacity of modular data centers, and related supernode instances will be launched in other regions one after another.