Overview
The HPC-AI Converged Computing Center at HKUST(GZ) integrates high-performance computing and AI acceleration capabilities, featuring industry-leading computational density, multi-node/multi-GPU collaboration, and unified heterogeneous resource management with elastic scalability. The center currently operates four major clusters (Phase I, II, III-ACD, EDA), delivering full-stack computing services for scientific research, AI training/inference scenarios.
All cluster resources are currently exclusively available to HKUST(GZ) faculty and students.
HPC Phase I has been operational since April 2022, delivering 0.246Pflops@FP64 double-precision computing capability and 5.597Pflops@FP16 half-precision AI computing power. The cluster consists of 12 Intel CPU computing nodes and 4 NVIDIA A30 GPU nodes interconnected via a 100Gb/s InfiniBand high-performance network, supported by a 701TB parallel file storage system.
HPC Phase II was officially launched in September 2023, comprising an international HPC-AI platform and a domestic AI platform. The HPC-AI platform provides 6.358Pflops@FP64 general computing capacity and 185.461Pflops@FP16 intelligent computing power through 146 Intel CPU nodes, 20 AMD CPU nodes, 65 NVIDIA A800 GPUs, and 15 NVIDIA A40 GPUs. The domestic AI platform focuses on localized computing with 19.040Pflops@FP16 capacity, equipped with 8 Atlas 300T Pro training nodes and 2 Atlas 300V Pro inference nodes. The entire cluster employs 200Gb/s InfiniBand networking and a 4.2PB hybrid storage architecture (309TB SSD + 3.9PB HDD).
HPC Phase III(ACD) commenced operation in January 2025, offering 18.933Pflops@FP64 double-precision computing capacity and 1,078.322Pflops@FP16 half-precision AI computing power. This cluster integrates 68 Advanced Computing Devices (ACD) nodes utilizing 400Gb/s RoCE v2 network protocol, complemented by a 17PB distributed storage system.
HPC EDA has been operational since July 2023, delivering 0.267Pflops@FP64 computational capacity and 5.682Pflops@FP16 AI acceleration capabilities. The cluster features 20 Intel CPU nodes and 4 NVIDIA A30 GPU nodes connected through a 200Gb/s InfiniBand network, supported by a 1.2PB parallel file storage system.