Atlas 800I A2-G

Atlas 800I A2-G Inference Server adopts 8-module efficient inference method, provides strong AI reasoning ability, and has advantages in computing power, memory bandwidth and interconnection ability. It can be widely used in large model reasoning, such as intelligent customer service, copy generation, knowledge precipitation and other content generation scenarios. NPU interconnection is supported to improve the inference efficiency of large models.

Technical Specification

Form	4U AI server
CPU	4 * Kunpeng920
Memory	32 DDR4 memory slots, up to 3200 MT/s
Memory	The capacity of a single memory bar supports 16/32/64 GB
Local Storage	8 * 2.5 SATA+2 * 2.5 NVMe
Local Storage	4 * 2.5 SATA+6 * 2.5 NVMe
RAID Support	Supports RAID 0/1/10/5/50/6/60
Internal network	Optional NPU fully interconnected model, the whole machine interconnection bandwidth 392GB/s
External network	8 * 200GE QSFP interface straight out, RoCE protocol
PCIe interface	Up to three PCIe 4.0 expansion slots are supported
Power Supply	Four hot-pluggable 2.6kW power modules support 2+2 redundancy
Power Supply	200 ~ 240V AC
Power Supply	240V DC
Heat dissipation mode	Air cooled
Fan	Support 8 hot plug fan modules, support N+1 redundancy
Operating Ambient Temperature	5℃～35℃（41℉～95℉）
Dimensions of Structure （H×W×D）	175mm × 447mm × 790mm

* The descriptions and information displayed in the product promotional materials are for reference only. The actual delivered product shall prevail. The final interpretation right belongs to GLORY.

Resources

Documents

Atlas 800I A2-G

Download