Introduction

HPC Phase III (ACD) Intelligent Computing Cluster comprises 68 computing nodes, each equipped with 8 GPU cards. The overall FP16 AI compute power of the cluster reaches 1051 Pflops. Each GPU card has a memory of 80GB. It employs advanced nanotechnology, offering a bidirectional bandwidth of up to 800GB/s between single GPU cards. For cluster interconnection, the ACD Intelligent Computing Cluster uses RoCE V2 networking technology, with a computing network speed of 400Gb/s and a storage network speed of 200Gb/s.

The ACD Intelligent Computing Cluster provides powerful computational support for faculty members, students, and research teams. It can be applied in various fields, including AI training and inference, scientific computing and research, data analysis and processing, virtual reality, and graphics rendering, thereby accelerating the progress of scientific research and teaching activities.