Building a solid foundation for independent scientific research, the SIXUNITED domestically developed information technology innovation server cluster has been deployed at Shanghai universities.
Release time:
2026-10-23
Computing demands in university research and teaching are steadily increasing, calling for high-performance computing clusters to support large‑model training and simulation experiments, while also necessitating domestically developed, secure‑and‑trustworthy infrastructure to ensure data sovereignty and controllability.
Recently, SIXUNITED completed the delivery of a domestically developed, indigenous innovation–based computing cluster project for a Shanghai university. The 64‑node indigenous‑innovation server cluster has officially been installed in the user’s data center, fully equipped with Zhaoxin Kaisheng® KH‑50000 domestic processors. Featuring a mixed architecture of 96‑core and 64‑core CPUs that can be dynamically scheduled as needed, the system delivers independently controllable hardware‑level computing power, providing high‑performance computational support for scientific research and teaching, AI training, big‑data analytics, and more, thereby advancing academic research breakthroughs and AI talent development.
01 The Power of China’s Indigenous Chips
This project employs the Zhaoxin Kaisheng® KH-50000 processor, balancing computational performance, security and reliability, and ecosystem compatibility, thereby providing independently controllable computing power to support diverse workloads such as data centers and cloud computing.
Self-reliant Core
It employs a self-developed microarchitecture and an in-house‑designed Chiplet interconnect, eliminating reliance on overseas IP from design through packaging. Having passed the National Security and Reliability Level II assessment, it establishes a solid foundation for independent and controllable operations.
The Core of Computing Power
Each individual chip supports up to 96 cores and up to 384 MB of L3 cache, enabling high‑concurrency, multi‑threaded computing. With 12 DDR5 ECC memory channels, total memory capacity can reach 3 TB, and the dual‑socket configuration delivers over 998 GB/s of bandwidth, providing ample memory resources for processing massive datasets and large models with hundreds of billions of parameters.
Shield of Safety
Equipped with the SM2, SM3, and SM4 national cryptographic instruction sets, as well as the SM9, ECS, and ZUC algorithms, it also supports security‑related technologies such as secure boot and key management, enabling end-to-end encryption and protection of business data across the entire stack—from the chip to the application.
02 Multi-Form Factor Computing Infrastructure
The domestically developed servers delivered this time cover three major scenarios: 2U storage servers, 2U general-purpose servers, and 4U AI computing servers, forming a versatile computing infrastructure that supports data storage, general-purpose computing, and AI workloads.
ZSA24V2-P AI Server
The ZSA24 V2 AI server, designed for AI computing workloads, features multi-GPU support and a flexible modular architecture, seamlessly handling the acceleration and scaling of compute-intensive tasks.
Supports dual Zhaoxin KaiSheng® KH-50000 processors, each with a TDP of 450W, and 24 channels of 5200MT/s DDR5 memory, ensuring high-speed read/write performance for massive scientific datasets. It also supports up to 10 dual-fan GPUs, with individual cards drawing up to 600W, and features 13 PCIe 5.0–compliant slots, fully meeting the high‑compute‑intensity requirements of university‑level large‑model training, parallel simulations, and other compute‑heavy workloads.
ZS22V2-P General-Purpose Server
The ZS22V2-P universal server is designed for cloud computing, database, and AI workloads, offering a compact form factor that seamlessly integrates compute, storage, and network scalability to meet the diverse configuration requirements of various business applications.
Equipped with dual KH-50000 processors and 24 DDR5 memory slots, it supports both 3.5‑inch and 2.5‑inch hard drives, with a maximum capacity of 28 drives. Fully compliant with PCIe 5.0 and backward compatible with CXL 2.0, the system features six FHHL PCIe expansion slots, two HHHL PCIe 5.0 slots, and two OCP 3.0 network interface cards. An integrated intelligent management chip delivers comprehensive remote, system‑level smart management capabilities. With flexible scalability, easy deployment and maintenance, it provides an all‑in‑one solution for building high‑performance computing infrastructure in university research environments.
03 Building Collaborative Clusters
This project, centered on the research and teaching needs of universities, integrates diverse hardware components into a versatile computing infrastructure that supports data storage, general-purpose computing, and AI workloads, thereby addressing the resource requirements of various applications.
The comprehensive multi‑form factor computing infrastructure is designed to precisely align power consumption and thermal management strategies with workload demands: storage servers and general‑purpose servers prioritize energy efficiency, while AI compute nodes are provisioned with ample scalability. Storage and compute servers adopt a decoupled architecture, enabling independent scaling as business needs evolve; homogeneous node clusters provide redundancy, ensuring uninterrupted operation of research workloads.

Server rack installation, power-on, and debugging.
In terms of specific model selection, two processor configurations are employed: a 96-core variant and a 64-core variant. The 96-core version features a base clock speed of 2.2 GHz paired with a large 384 MB cache, optimized for high-concurrency, multi-threaded workloads. The 64-core version, on the other hand, excels with a higher base clock of 2.6 GHz, delivering strong single-core performance while consuming less power. By aligning node‑level task requirements with memory, storage, power consumption, and thermal management needs, resource allocation is tailored to better match real‑world workloads.
This solution both avoids resource underutilization caused by a one-time large-scale investment and enables data to flow locally within the cluster, thereby closing the end-to-end loop from data accumulation to intelligent applications.
04 Standardized Delivery
The implementation of a computing‑power cluster solution requires that equipment configurations be meticulously aligned with every stage of the data‑center deployment process.
In this project, the delivery team adhered rigorously to standardized procedures, completing all stages—including equipment receipt, unpacking and inspection, rack mounting, structured cabling, and power‑on commissioning—thereby advancing the on‑site deployment of a computing cluster comprising 64 domestically developed servers. This ensured the orderly handover of equipment and laid the groundwork for subsequent business operations.
Real-World Server Deployment and Cabling
Targeted at university research and teaching, this computing cluster supports large‑model inference, multimodal research, AIGC experiments, as well as tasks such as finite‑element simulations, bioinformatics analysis, and batch data cleaning. At the same time, it can serve as a platform for adapting and validating domestically developed software and hardware, enabling compatibility testing and stress validation of domestic databases, AI frameworks, and enterprise systems.
In addition, this type of multi‑form factor computing solution can also serve hybrid government‑and‑enterprise business scenarios, supporting applications such as office automation (OA), unified identity management, and offline big data analytics, thereby meeting the diverse computational requirements of business systems, data warehouses, and intelligent applications.
The successful delivery of this university-level indigenous innovation computing cluster project marks another milestone for SIXUNITED in the indigenous innovation computing field. By coordinating domestically produced processors with multi-form-factor servers, SIXUNITED provides robust computational support for university research and teaching, while also demonstrating its capability to tailor products and deliver projects that meet real-world business needs.
The domestic computing infrastructure has entered university server rooms, marking not only the deployment of hardware equipment but also laying a solid foundation for data autonomy and controllability in research and teaching. Riding the wave of digital transformation in education, SIXUNITED will continue to refine its domestically developed computing hardware, leveraging its robust domestic computing capabilities to bolster China’s self-reliant and self-strengthening research infrastructure.
recommend News