6
2 Product Specifications
2.1 Overview
Inspur AI server NF5488A5 has high scalability, high performance, high energy efficiency,
flexible deployment and other features. Its AI computing performance can reach 2
petaflops, suitable for image & video, speech recognition, financial analysis, intelligent
customer service and other AI application scenarios. With the massive growth of data and
the rapid iteration of models, AI research institutes and commercial companies urgently
need to improve AI computing power to shorten the model training and development cycle.
At the same time, they hope to deploy AI infrastructure more quickly and economically, and
to realize the compatibility between the AI infrastructure and the legacy IT infrastructure
to save data center space and reduce costs. NF5488A5 uses the most advanced NVIDIA
NVSwitch interconnect architecture in the industry,and can be equipped with eight NVIDIA
SXM4 A100 Tensor Core 40GB/80GB GPUs interconnected at high speed in 4U space.The
direct P2P data interaction between any two GPUs can achieve 2 petaflops of AI computing
performance.
. NF5488A5 adopts the most advanced high-speed NVIDIA NVSwitch™, which
enables P2P connection between any two of the 8 NVIDIA SXM4 A100 Tensor Core 40/80 G
GPUs at a total bandwidth of up to 600 GB/s in a 4U space. With 2 AMD EPYC 7002/7003
series PCIe 4.0 CPUs, together with xGMI-2, this server provides top-level computing
performance. A 4U chassis and power supply redundancy design enable NF5488A5 to be
widely applied to data center environments, especially mounted to the cabinets with
limited power consumption. Besides, NF5488A5 adopts a more flexible cluster deployment
scheme for integration from hardware to applications. Moreover, the 54V_VR power supply
offers higher power efficiency. A combination of the layered and zoned cooling channels
and an intelligent PID control strategy ensures optimal cooling performance. With
NF5488A5, AI users can build AI infrastructures and development environments efficiently
with high computing performance and low deployment and operational costs.
