4U 8x A100 GPU Server for AI Training and Inference

Versatile 8GPU Server for AI Workloads

Ranking first in single sever performance in MLPerf 0.7, it supports 8 fully interconnected NVLink 3.0 NVIDIA A100 Tensor Core GPUs, and 2 AMD® EPYC® Rome processors enabling PCIe4.0 in a 4U space, NF5488A5 delivers outstanding AI performance of 5 petaFLOPS and ultra-high data throughput, making it ideal choice for AI applications like image, video and voice processing, financial analysis and virtual assistants.

Key Features

Record-Setting AI Performance

Best single sever performance in the MLPerf v0.7 benchmark, setting 18 performance records and achieving 549,000 inferences per second in the ResNet50 inference task.

Learn more: Press Release | Blog Post | Official MLPerf v0.7 Results


Extreme Hardware Design

8x fully interconnected NVLink 3.0 NVIDIA A100 GPUs, 2x AMD EPYC Rome processors enabling PCIe4.0 in 4U and delivering 5 petaFLOPS and high data throughput.


Accelerating AI Training

Supports the latest A100 80G GPU with HBM2e, doubling the memory capacity of its predecessor and a 30% increase in GPU memory bandwidth in a single node, delivering 15% performance increase in AI model training tasks with tens of billions of parameters.

Model NF5488A5
GPU 1* HGX  A100 8GPU
Processor 2* AMD® EPYC® Rome Processor (Up to 225W TDP)
Performance 5 petaFLOPS
Memory 32* DDR4 2933MT/s RDIMMs//LRDIMM, up to 4 TB
Storage 8* 2.5” SAS/SATA (or 4*NVMe+4* SATA/SAS), 4* NVMe M.2, 2* SATA M.2
PCIe 4* built-in low profile PCIe x16
Front I/O 2* USB 3.0 port, 1* VGA port, 1* RJ45 management port
Ethernet Card 10G Ethernet Optical Interface
Management Built-in Aspeed2500 BMC module, support IPMI, SOL, KVM Over IP, VM, etc.
RAID Support 3508 Raid Card (NVMe Raid with Raid Key)
OS Red Hat Enterprise 7.8 64bit, CentOS 7.8; Ubuntu 18.04
Cooling N+1 Redundant hot swap fans
PSU 3+1 Redundant PSUs
Chassis 4U, 448mm W x 175.5mm H x 850mm D