Product Introduction
KunLun G5680 V2 is a 4U 4-socket AI training and inference server based on Kunpeng 920 and Ascend 910 processors. It features supreme computing density, ultra-high energy efficiency, and high-speed network bandwidth. KunLun G5680 V2 is widely used for deep learning model development and training and is ideal for industries that require high computing such as smart city, smart finance, education, scientific research, and telecoms.
Supreme Computing Density
4 x Kunpeng 920 processors
Ascend 910 AI processors
Ultra-High Energy Efficiency
AI computing infrastructure featuring high energy efficiency and low power consumption
Kunpeng 920 processors with a 30% higher energy efficiency ratio compared to industry benchmarks
High-Speed Network Bandwidth
8 x 200 Gbit/s RoCE v2 high-speed ports
Interconnection of 8-socket NPUs through HCCS full mesh with a bidirectional bandwidth of 392 GB/s
Technical Specifications
4U AI training and inference server
4 x Kunpeng 920 processors (48/64 cores), with a CPU frequency of 2.6 GHz/3.0 GHz
Up to 32 x DDR4 RDIMMs, up to 3200 MT/s, and up to 2 TB memory capacity
Ascend 910 AI processors
7 x 2.5″ SATA drives + 3 x 2.5″ NVMe SSDs
8 x 2.5″ SATA drives + 2 x 2.5″ NVMe SSDs
4 x 2.5″ SATA drives + 6 x 2.5″ NVMe SSDs
RAID 0, 1, 10, 5, 50, 6, and 60
8 x pass-through 200GE QSFP ports, supporting the RoCE protocol
Up to 3 x PCIe 4.0 expansion slots
8 x hot-swappable fan modules in N+1 redundancy
The BMC chip integrates one dedicated management GE network port, providing comprehensive management features such as fault diagnosis, automatic O&M, and hardware security hardening.
The BMC supports standard interfaces such as Redfish, SNMP, and IPMI 2.0, provides a remote management interface based on HTML5/VNC KVM, and supports out-of-band management functions such as monitoring, diagnosis, configuration, Agentless, and remote control for simplified management.
It is optional to configure the management software that provides advanced management features such as stateless computing, batch OS deployment, and automatic firmware upgrade, realizing an intelligent and automatic management throughout the lifecycle.
4 x hot-swappable 2600 W or 3000 W AC Platinum/Titanium PSUs in 2+2 redundancy (configured based on the power consumption of the server and the power of a single PSU)
FusionOS, BCLinux, SLES, Ubuntu, CentOS, UOS, and Kylin
5°C to 35°C (41°F to 95°F), compliant with ASHRAE Classes A1 and A2
175 mm x 447 mm x 790 mm (6.89 in. x 17.60 in. x 31.10 in.)
Copyright © 2025 Kunlun Technologies International Co., Limited All rights reserved.