IEIT Systems NF5468G7 — 8 to 10 GPU AI Server (Inspur)

Specifications

Overview

The NF5468G7 is the current flagship of IEIT Systems' NF5468 AI server series — the line sold internationally under the Inspur meta brain brand. It is a 4U, dual-processor, eight-to-ten GPU platform offered in two hardware variants: an Intel configuration on two 4th Generation Xeon Scalable processors with up to 112 cores and eight dual-slot GPU cards, and an AMD configuration on two 5 nm 4th Generation EPYC processors with up to 192 cores and ten full-height, full-length, double-width GPU cards.

Both variants make the same architectural bets. The PCIe 5.0 fabric supports accelerators up to 600 W, CXL 1.1 is supported for memory space expansion and memory sharing across the system, and networking is provisioned for 400G NDR InfiniBand alongside a 400G OCP 3.0 mezzanine slot. The AMD variant uses a ten-card CPU-GPU pass-through design to cut communication latency and raise GPU utilisation, backed by up to sixteen NVMe or E3.S drives; the Intel variant leans on memory and storage density instead, with up to 8 TB of DDR5 across 32 DIMM slots and 300 TB of local high-speed storage.

Cooling and serviceability follow the same pattern: N+1 redundant fans with a claimed 25% cooling improvement and 5% noise reduction, plus four 1600 W to 3000 W Platinum or Titanium power supplies in N+N redundancy. IEIT pairs the hardware with its AIStation resource platform and a three-level hardware, firmware and software security framework, which is how the platform is positioned for AIGC, deep learning, AI for Science and metaverse workloads.

Key Benefits

Two platforms, one chassis — Intel Xeon for memory and storage density, AMD EPYC for maximum core count and a ten-card pass-through topology. Current interconnect — PCIe 5.0 with 600 W accelerator support, CXL 1.1 memory expansion and 400G class networking. Density at scale — up to 8 TB of DDR5 and 300 TB local storage in the Intel variant, sixteen NVMe/E3.S drives in the AMD variant. Fleet-friendly serviceability — N+N power, N+1 fans and a 25% cooling improvement with reduced acoustic output.

Applications

Deep learning training and large-model inference, AIGC and generative media pipelines, AI for Science and HPC-adjacent simulation, metaverse and real-time rendering workloads, and enterprise or carrier AI clusters built around 400G InfiniBand or RoCE fabrics.

Related Products