Product Positioning: Ultimate Density, Liquid-Cooled Cooling, the Preferred Choice for Large Model Training Clusters. A high-density computing power solution specifically designed for next-generation intelligent computing centers, addressing the core pain points of insufficient computing power, excessive energy consumption, and limited data center space in large model training through exceptional computing power density, ultra-low PUE, and rapid scalability.

core strengths

Ultimate increase in computing power density:

The computing power density of a single cabinet is three times higher than that of traditional air-cooled solutions, saving 50% of the server room space under the same computing power demand, significantly reducing data center land, infrastructure, and cabinet rental costs, adapting to the large-scale deployment of high-density computing power clusters, doubling the computing power capacity in limited server room space, and easily supporting large-scale model training and high concurrency computing tasks.

Full scene liquid cooling and heat dissipation technology:

Supports two liquid cooling solutions: cold plate and immersion, with flexible selection and heat dissipation according to the scene

The efficiency far exceeds traditional air cooling, which can stably control the temperature of core components within the optimal operating range, avoiding high load frequency reduction; The PUE of the entire machine is as low as 1.1 or below, saving more than 50% energy compared to traditional air cooling solutions, significantly reducing long-term electricity costs, helping data centers transition to green and low-carbon, and meeting the requirements of energy conservation, consumption reduction, and sustainable development.

Ultra high GPU integration capability:

8-16 high-end AI acceleration cards can be integrated in a compact 5U space, with full computing power on a single node, eliminating the need for multi node split deployment and simplifying the cluster topology architecture,

Reduce cross node data exchange losses,

Reduce network interconnection costs and failure points, fully meet the ultimate computing power requirements of multi billion parameter large model training, massive data parallel computing, and ultra large scale inference clusters.

Minute level elastic expansion capability:

Fully modular hot swappable design, computing power nodes, liquid cooling modules, and power supply units can be independently and quickly replaced and expanded without the need for complete machine shutdown. The computing power upgrade can be completed in minutes, quickly responding to business peaks and fluctuations in computing power demand, achieving on-demand scheduling and elastic scaling of computing power resources, and ensuring continuous and stable business operation.

Unified control and high reliability:

Built in intelligent monitoring system, real-time monitoring of key indicators such as temperature, flow rate, power consumption, and computing load, supporting remote centralized management and fault warning; The whole machine adopts a highly reliable industrial grade design, which is resistant to vibration and operates in a wide temperature range. It is suitable for 24/7 uninterrupted operation in large data centers, ensuring uninterrupted training tasks and data loss.