AI Training Cluster and Data Center Inspection
Internet Data Center

AI Training Cluster and Data Center Inspection

Project Background

Customer added training workloads, but air-cooled cabinets lack sufficient power density. Mixed training and inference workloads result in low resource utilization. Large data center footprint creates staffing challenges for night patrols.

Solutions

Training nodes are delivered as full racks. The high-power zone uses liquid cooling, while the standard zone uses air cooling. NVIDIA GPUs and Supermicro servers arrive and go online in batches. For networking, Huawei OTN and Ruijie switches handle east-west traffic, with Inspur servers managing inference and scheduling. In the data center aisles, Lunqu wheeled inspection robots collect indicator light and environmental data, triggering alert tickets when anomalies are detected.

Use Device

NVIDIA GPUs, Supermicro GPU Servers, Inspur Servers, Huawei Transmission Equipment, Ruijie Switching Equipment, Lunqu Wheeled Chassis

Implementation Results

Training and inference are deployed in separate zones, enabling unified scheduling of idle resources. Facility conditions and equipment status are collected regularly by inspection devices; anomalies automatically generate work orders, reducing the frequency of manual night inspections.

Have a similar project need?

Get in touch now

INDUSTRY SOLUTIONS

Internet Data CenterIndustry Solutions

This case belongs to the Internet Data Center industry.Charge for compute, networking, and operations per rack.. The industry includes a total of3Solution packages can be reused or combined based on project scope.

View all solutions for this industry
Beijing ICP License No. 2026026760