Description of the event
On May 17, the 2024 AI Empowering Enterprise Business Innovation Forum with the title “Breaking Business Boundaries, Evolving the Way” came to an end. At this forum, Shenzhou Kuntai, a subsidiary of Shenzhou Data, released a variety of products and solutions including the heterogeneous intelligent computing scheduling and operation platform HISO, the heterogeneous intelligent computing acceleration platform HICA, and the integrated delivery of Shenzhou Kuntai full-liquid cooling cabinet products to help enterprise customers land green in the entire multi-cloud heterogeneous environment Matching in heterogeneous intelligent computing resource pools A better combination of computing power, effectively improving the resource usage efficiency of GPU server clusters, and effectively solving the energy consumption problems of nodes and inter-node connections.
Incident comments
In response to the energy consumption problem of interconnection between nodes, silicon light+liquid cooling creates a new energy saving solution. Shenzhou Kuntai uses silicon light technology to reduce modulator voltage through multiple modulators with a single light source, and uses a series of technologies such as distributed feedback lasers, which can effectively reduce interconnection energy consumption by 25%. At the same time, Shenzhou Kuntai's liquid cooling server improves the cooling efficiency of the system through integrated cold plates and intelligent flow regulation, and uses negative pressure piping systems and near-end liquid leakage detection technology to improve the reliability of the cooling system, thereby effectively reducing node energy consumption by 30%. The silicon light+liquid cooling solution enabled Shenzhou Kuntai's entire cabinet product to be fully upgraded, effectively solving the high energy consumption problem of interconnection between nodes.
HISO, the heterogeneous intelligent computing scheduling and operation platform, and HICA, the heterogeneous intelligent computing acceleration platform, effectively improve the utilization of computing power resources. In the process of large model training, the low utilization rate of computing power resources greatly increased the usage costs of enterprises. According to relevant data, the MFU (Model Flops Utility) of OpenAI training GPT-4 is between 32% and 36%, while the average MFU utilization rate in the industry is currently only 30% to 40%. Based on cloud-native technology, HISO integrates GPU hard sharding and virtual sharding technology to virtualize or pool GPU resources, thereby completing computing power scheduling across clusters. HICA, on the other hand, is expected to reduce idle computing power by 20% by blocking differences in the underlying computing power ecosystem within the cluster, and effectively decomposing parallel computing tasks and matching the corresponding software stack and computing power resources.
Integrated delivery is easier to deploy and effectively reduces the complexity of deployment and operation and maintenance. At the opening ceremony of the Shuyun Force Conference 2024, Shenzhou Kuntai officially released the “KunTai Pod2000 Full Liquid Cooling Cabinet” solution product. The solution uses integrated delivery to help customers avoid problems such as multiple interfaces, complicated joints, difficult on-site implementation and deployment, and long implementation cycles.
At the same time, the whole cabinet solution realizes 100% full liquid cooling and a cost-effective liquid cooling solution, which will effectively control energy consumption costs for users.
The new products solve users' pain points, and product sales are expected to increase. Products such as HISO, the heterogeneous intelligent computing scheduling and operation platform HISO, the heterogeneous intelligent computing acceleration platform HICA, and the integrated delivery of Shenzhou Kuntai liquid cooling cabinet products will effectively solve the problems of high energy consumption, low resource utilization, and deployment, operation and maintenance encountered by users in large model training and large computing power requirements, thus driving the company's server product sales. The company's net profit for 2024-2026 is estimated to be 14.2 billion yuan, 1.7 billion yuan, and 2 billion yuan, respectively. The current market value corresponds to PE at 17, 14, and 12 times, respectively, maintaining a “buy” rating.
Risk warning
1. The risk of increased market competition;
2. Risk of supply chain shortages;
3. Demand for computing power falls short of expectations;
4. The pace of liquid cooling landing falls short of expectations.