Source: H3C |Original Link

Currently, the large-scale deployment of AI foundation models has driven a surge in demand for intelligent computing power. Domestic computing hardware characterized by high density, high reliability, and secure controllability has become a critical requirement for the construction and upgrading of computing clusters across various industries. To precisely address industry compliance requirements and AI computing challenges, H3C Group, a subsidiary of Unisplendor Corporation, has launched the H3C UniServer R4930 G7 2U domestic compute-intensive server. As the only domestic 2U model capable of supporting four 600W high-power AI accelerator cards at full configuration, this system satisfies both policy compliance and the training and inference demands of large models, thereby establishing a robust foundation for autonomous intelligent computing within the industry.

Surging Industrial Demand: Domestic Computing Power Enters a Window for Large-Scale Deployment

Industry monitoring data from IDC indicates that in 2025, domestic hardware accounted for 41% of AI accelerator card shipments in China, with the growth rate of the local computing hardware market significantly exceeding the industry average. By the end of 2025, the overall domestic market share of domestic servers surpassed 30%, as procurement volumes of localized equipment continued to rise in key sectors including finance, telecommunications, and government and enterprise services. Driven by the exponential increase in computing consumption resulting from the widespread adoption of large models and AI agents, high-density, full-stack secure and controllable domestic computing servers have become the primary choice for new deployments and capacity expansions across industries.

Proprietary Self-Developed Architecture: Flagship AI Performance in a 2U Form Factor

Most domestic 2U servers currently available support accelerator cards up to 350W–400W only, making them insufficient for next-generation large model workloads. The R4930 G7 achieves a breakthrough through its proprietary power supply and cooling architecture:

Stable Operation with Four 600W Cards at Full Configuration:Featuring independent redundant power delivery for GPUs and dedicated high-power modules, this system offers a unique capability among domestic devices of the same specification to run four 600W double-width accelerator cards in parallel at full load. It ensures sustained operation without frequency throttling or card disconnection, making it suitable for specialized industrial AI training and large model inference. Additionally, it flexibly supports six full-bandwidth single-width cards or up to 14 lightweight accelerator cards, ensuring compatibility with mainstream domestic AI hardware.

Ultimate Computing Density:Housed in a standard 2U air-cooled rack chassis, the system supports up to 128-core domestic processors and handles hybrid workloads comprising general-purpose computing and AI tasks. It is particularly well-suited for scenarios such as inference for ten-billion-parameter large models (e.g., product recommendations, knowledge Q&A, short-form content generation) and multi-instance operations on a single machine (e.g., cloud desktops, graphics rendering), significantly conserving rack space and reducing cluster construction and O&M costs.

PCIe 5.0 High-Speed Expansion:The system is equipped with 14 PCIe 5.0 slots and two OCP 3.0 network adapters, providing ultra-high bandwidth that effectively eliminates I/O bottlenecks in large model data transmission.

Elastic Modular Storage Options for Diverse Data Workloads

This model offers flexible storage configurations that can be customized on demand to meet requirements for high-speed read/write operations and massive data storage. It features eight SFF/NVMe high-speed drive bays paired with four LFF high-capacity drives, with optional M.2 local caching. The system supports multi-specification hard drive modules and allows for the flexible combination of computing, storage, and network resources, making it suitable for scenarios such as distributed storage, video analytics, model caching, and database applications.

100% Domestically Produced System with End-to-End National Cryptographic Security Protection

The system achieves a 100% domestic production rate for core components, including a fully domestically produced BMC management board, fully meeting the stringent procurement standards of central and state-owned enterprises and the financial sector. It provides comprehensive underlying security capabilities: the CPU integrates a hardware-based trusted security module and natively incorporates hardware acceleration engines for the complete suite of national cryptographic algorithms (SM2, SM3, and SM4). This enables memory encryption, robust virtual machine isolation, and end-to-end storage encryption, establishing a foundational hardware defense barrier that effectively resists various external physical intrusion attacks. Furthermore, all BIOS and BMC firmware are independently developed with complete proprietary intellectual property rights, thereby mitigating risks associated with underlying security vulnerabilities.

Full-Stack Domestic Ecosystem Compatibility to Reduce Business Migration Costs

The product has completed full-chain certification for domestic software and hardware, significantly reducing adaptation investment. It is compatible with mainstream domestic operating systems, middleware, and other foundational software, as well as major domestic cloud platforms, enabling the seamless migration of office workloads, business applications, and AI innovation initiatives. Covering eight key sectors—including government affairs, finance, internet, telecommunications, and electric power—it supports mainstream digital core scenarios such as government large models, financial risk control, cloud-network operations, energy dispatching, urban visual analytics, and enterprise private intelligent computing.

As a leader in digital and AI solutions, H3C adheres to the philosophy of "Built-in Intelligence, Empowering Smart Computing" to refine its full-stack portfolio of domestically produced computing products. The R4930 G7 addresses the gap in high-density, domestically produced AI computing power, assisting government and enterprise clients in meeting localization substitution targets while providing a stable and reliable infrastructure foundation for domestic large models and industry-specific intelligent applications.