云工场科技 - 引领边缘计算与 AI 融合

The Nation's First AMD

ROCm on Radeon Open Source Ecosystem Intelligent Computing Center

Building the nation's largest AMD AI computing engine closest to advanced manufacturing customer clusters

Computing Power & Servers

Phase I project investment of 240 million, providing 5000 cards 615P AMD Radeon W7900D series, empowering thousands of industries for AI+ scenario innovation and large models, providing API (tokens) to bare metal inference computing resources and services.

Card Configuration & Performance

Each compute unit equipped with 2nd gen Ray Tracing Accelerators and AI Accelerators

Based on AMD RDNA 3 architecture, chiplet design

48 GB

Large VRAM, single card supports multi-model inference

295 w

Total Board Power (TBP)

91 TELOPs

Half Precision (FP16) Performance

864 GB/s

Peak Memory Bandwidth

Consult Now

AMD Radeon W7900D

Node Configuration & Performance

4 Nodes Stable Operation

DeepSeek 671B Full Model

GPU

AMD Radeon W7900D Series * 8

CPU

EPYC™9334 3.9 GHz 32 Cores * 2

Data Disk

Hippstor 7.68TB * 1

System Disk

Samsung SATA 480GB * 1

NIC

MellonaX Dual Port 25G_CX5 * 2

Chassis

Consult Now

Free Test Computing Power

And Series Special Support

Up to 99.99% Reliability Cluster Resources

From Test Verification, Migration Optimization

To Model Fine-tuning, Algorithm Development Full Chain Service

7x24 Hour Professional Support Team

Plus AMD Original Factory Technical Resources

Facility Conditions

Basic Conditions: Dual-route external mains access, ring network water supply guarantee, 4-route fiber optic office direction.

Location

Wuxi Mashan Computing Island

One of Mobile 3+3+X Strategic Deployment Data Centers

Capacity

Cluster contains 200+ high-power racks

6-15KW Elastic Expansion

Green

Annual PUE as low as 1.248

Meets International T3+ Standards

Access Advantages

Based on All-Optical Mesh OXC Technology

Create low latency, ultra-large bandwidth, ultra-high reliability transmission links

Direct Access to National Internet Backbone Direct Connection Points

Local computing power quickly merges into the national computing network, accurately delivering computing data

3ms Latency within Reach

Located in the geometric center of the Yangtze River Delta
Computing power reachable in one hop, ensuring 3ms latency within reach

Target Audience

Empowering Internet, Finance, Cultural & Creative Education, Manufacturing Energy, Medical Research and other fields, assisting in large-scale, multi-modal real-time inference, real-time simulation, model algorithm development and other scenarios, building open decoupled, technology autonomous and controllable computing infrastructure. Serving JD.com, 58.com, Shenzhen University, China Digital Energy, Kexuntong