Alibaba Cloud Lingjun Zhenwu M890 Supernode Instance Launches 64-GPU AI Computing Platform

Alibaba Cloud introduces advanced AI supernode computing platform

Alibaba Cloud, the cloud computing division of Alibaba Group Holding Limited, has announced the launch of the Alibaba Cloud Lingjun Zhenwu M890 Supernode Instance, a high-speed interconnected 64-GPU computing unit based on a supernode architecture. The platform is currently available through an invitation-only preview in the Ulanqab region and is designed to support demanding artificial intelligence workloads, including autonomous driving and embodied AI applications.

The Lingjun Zhenwu M890 supports FP8 and FP4 low-precision computing capabilities, enabling efficient processing for large-scale AI model development. By using the ICN Switch 1.0 chip, the system expands scale-up GPU interconnect capability from 16 GPUs to 64 GPUs while increasing GPU-to-GPU interconnect bandwidth to 800 GB/s.

For AI training workloads, including autonomous driving scenarios, the new architecture delivers three times the training performance compared with the previous-generation Zhenwu 810E platform. A single supernode instance can support inference operations for Mixture-of-Experts large language models with up to 10 trillion parameters, enabling organizations to manage increasingly complex AI models.

The platform is part of broader AI infrastructure development by China-based technology companies seeking higher computing efficiency and scalability. Alibaba Cloud is focusing on advanced computing architectures that can support future AI applications requiring extensive processing resources, faster data movement, and improved model deployment capabilities.

High-scale GPU networking architecture for AI workloads

The HPN 8.0 integrated training and inference network architecture supports heterogeneous deployment of up to 130,000 GPUs within a single cluster. The architecture can be further expanded beyond one million GPUs, providing a foundation for large-scale artificial intelligence computing environments and supporting future developments in enterprise AI applications.

The combination of high-bandwidth GPU connectivity, low-precision computing support, and scalable cluster architecture positions the platform for advanced AI training and inference use cases. These capabilities are particularly relevant for sectors requiring intensive computational performance, including autonomous systems, large language models, and embodied intelligence solutions.

Frequently Asked Questions

What is the Alibaba Cloud Lingjun Zhenwu M890 Supernode Instance?
The Alibaba Cloud Lingjun Zhenwu M890 Supernode Instance is a high-performance computing platform designed to provide a 64-GPU interconnected architecture for advanced artificial intelligence workloads. It supports FP8 and FP4 low-precision computing, increases GPU communication bandwidth, and enables large-scale AI model training and inference. The system is designed for applications such as autonomous driving, embodied AI, and large language model processing with extremely high computational requirements.