Empowering Research, Creating an Intelligent Future

Access portal:https://ai4s-cloud.pkusz.edu.cn/

The AI4S High-Performance Computing Platform (hereinafter referred to as “the Platform”) of Peking University Shenzhen Graduate School is a strategic infrastructure oriented toward intelligent science and interdisciplinary research. It was officially completed in 2025 and opened to the entire institute. Adhering to the concept of “computing power drives innovation,” the Platform is committed to building a research engine integrating high-performance computing, intelligent algorithm optimization, and multidisciplinary collaborative R&D, providing foundational computing power support for artificial intelligence, scientific intelligence, and engineering technology innovation at the graduate school.

The Platform adopts an industry-leading heterogeneous computing architecture, with the following specific hardware configurations:

Large-scale parallel computing servers: Each equipped with 8 A800 GPUs, optimized for large-scale AI model training and high-performance scientific computing tasks;

Extreme-performance computing servers: Each equipped with 8 A100 GPUs, designed for cutting-edge scientific research and complex numerical simulations with extreme computing demands;

General-purpose computing workstations: Provide cost-effective local computing support for small- and medium-scale AI application development, algorithm prototyping, and teaching demonstration scenarios;

High-performance storage system: Configured with a 300 TB all-flash storage array supporting NVMe protocol, ensuring high-speed data read/write and high-concurrency access capabilities.

In terms of computing performance, the Platform provides:

General computing capacity: 350 TFLOPS (trillions of floating-point operations per second)

AI-specific computing power: 12.5 PFLOPS (quadrillions of floating-point operations per second)

Distributed parallel computing: Supports high-speed interconnection between nodes, enabling large-scale model parallel training across servers

Platform service matrix:

Public Computing Power Services (Computing Resource Scheduling)

Core functions: Achieves dynamic allocation of multi-task priorities and concurrent multi-user processing through an intelligent scheduling system

Usage method: Users submit computing tasks through a unified web interface, and the system automatically allocates optimal computing resources

Applicable scenarios: Short-term computing needs such as small- and medium-scale AI model training, scientific simulation, and algorithm validation

Service advantages: Supports full-process services from basic computing support to professional technical guidance, enabling on-demand usage and cost optimization

Dedicated Computing Power Services

Core functions: Provide exclusive access to entire servers, ensuring complete isolation and stability of computing resources

Optional configurations: Support full rental of A800 servers, A100 servers, and RTX 5090 workstations

Applicable scenarios: Long-term large-scale model training, research projects involving sensitive data, and tasks with special computing environment requirements

Service advantages: Guarantee zero interference in computing resources, with customized operation and maintenance support and performance optimization suggestions

Hosting Services

Core functions: Provide professional data center hosting services for research teams’ own computing equipment

Hosting environment: Enjoy the same standards as the platform’s native equipment, including power supply, cooling systems, network access, and security protection

Applicable scenarios: Deployment of specialized computing equipment by research teams, industry collaboration project equipment deployment, and customized hardware requirements

Service advantages: Eliminate the cost of building and maintaining private machine rooms, while enjoying a professional-grade data center environment and 24/7 technical support

The Platform supports research directions including artificial intelligence and deep learning algorithm development, scientific intelligent computing and multi-physics simulation, big data analysis, and interdisciplinary modeling. In the future, the Platform will continue to expand its computing node clusters, explore the integration and application of emerging technologies, and deepen industry–academia–research collaboration. Through a comprehensive service system and advanced computing resources, it will provide sustained momentum for research innovation at the graduate school and support frontier exploration in intelligent technologies.

Faculty and students across the institute are sincerely invited to apply for access to the AI4S High-Performance Computing Platform and jointly explore the frontiers of intelligent technology.