KnowledgeNet Domestic Compute Service Project, Full-stack Domestication, Building Digital AI Compute Foundation
Adopting a flexible procurement model of framework agreement plus purchase order, based on the full-stack domestic technology route of Kunpeng+Ascend, building an independent and controllable AI computing infrastructure for CNKI digital publishing business, with model optimization and integration services, promoting the upgrading and development of intelligent publishing.
Background and Challenges
CNKI as a leading digital publishing platform in China is seeing increasing demand in AI applications such as knowledge graph construction, intelligent retrieval, content generation, semantic understanding, and literature analysis. There is an urgent need for high-performance AI computing infrastructure. However, traditional imported computing solutions pose supply chain security risks, and technological independence and controllability have become a major concern. At the same time, the procurement of AI computing power faces a dilemma: large-scale one-time investment leads to resource闲置, while scattered procurement is difficult to ensure规范性 and price advantages. How to ensure procurement flexibility while achieving full-stack domestic substitution is a key issue for CNKI's digital transformation.
-
requirement of independent and controllable technology
There is supply chain uncertainty in the imported computing power solution, and the full-stack domestic substitution is a hard requirement of the national strategic of information creation
-
insufficient procurement flexibility
One-time large-scale procurement causes resource idleness, and scattered procurement is difficult to ensure standardization and price advantage
-
growth in AI computing power demand
The demand for AI computing power continues to grow in intelligent search, content generation and other AI applications, and high-performance training and inference capabilities are required
-
missing landing service
It is difficult to directly support AI applications with only hardware procurement, and lacks full-process technical support from deployment to optimization
Solution
Adopting a flexible procurement model with framework agreements and purchase orders, both parties first sign a framework agreement to agree on technical standards and benchmark prices, and then issue purchase orders in batches according to business needs. Based on the full-stack nationalized technology route of Kunpeng CPU + Ascend NPU +银河麒麟OS, with AI model optimization and network integration value-added services, a total of 4 nationalized AI servers have been delivered in two batches, providing autonomous and controllable high-performance AI computing power support for the digital publishing business of Zhiwang.
AGREEMENT — Framework Agreement Signing
Agree on the overall cooperation terms; clarify technical standards and benchmark prices; stipulate quality requirements and acceptance procedures; determine the liability system for breach of contract
DELIVERY — Delivery of Batch Orders
First batch: 2 integrated machine servers; second batch: 2 servers + value-added services; Ascend 910 FP16 312.5 TFLOPS; 200GE ROCE high-speed interconnection
SERVICE — Integrated Service Implementation
AI model planning, design and optimization; network implementation planning and integration; three-level acceptance quality assurance; 5 years of original factory warranty service
Core characteristics
Full-stack localization, benchmark practice of information technology innovation
A full-stack localization solution using Kunpeng processors, Ascend AI chips, and the银河麒麟 operating system, achieving complete localization substitution from CPU, NPU to the operating system, which is a typical information technology innovation project and conforms to the development direction of national information technology application innovation.
Framework procurement model, flexible order as needed
Adopting a framework agreement plus procurement order model, the buyer flexibly procures in batches according to the needs of business development, ensuring procurement standardization and price advantage while avoiding resource waste caused by one-time large-scale investment, and accurately matching the rhythm of business development.
High-performance AI computing power, driven by Ascend 910
Adopting the high-performance Ascend 910 series AI chips, the single-card FP16 computing power reaches 312.5 TFLOPS, with the support of a 200GE ROCE high-speed interconnect network, it supports large-scale distributed AI model training and meets the computing needs of complex AI tasks such as large language models and multimodal models.
Integrated service delivery, full-process quality assurance
Complementary AI model planning and design optimization, network implementation planning and integration, technical training and other value-added services, establish a three-level acceptance process of delivery acceptance, initial project inspection, and final project inspection, 5-year original factory warranty and lifetime quality responsibility commitment, providing full-process protection from hardware to application for customers
Application Scenario Display
Profit and value
Build an autonomous and controllable AI computing power foundation with full-stack nationalization to empower the intelligent upgrade of digital publishing.