Inspur / IEIT NF5468M6 AI Server Datasheet

The NF5468M6 is a 4U versatile AI server optimized for artificial intelligence training, high-density inference, intelligent video analytics (IVA), and graphics rendering. Built on the 3rd Gen Intel® Xeon® Scalable platform, it features dynamic topology switching (Balance, Common, Cascade) and extensive accelerator card compatibility.

Product Variants

• NF5468M6-P (Mainstream Adaptive): Designed for mainstream AI training and inference with up to 8 double-width PCIe GPUs and software-selectable topology modes.

• NF5468M6-T (Cost-Effective Pass-Through): Optimized for lightweight AI training/inference using 4 double-width GPUs with a CPU-GPU direct pass-through topology (no PCIe switches) to eliminate latency and reduce TCO.

• NF5468M6-V (High-Throughput / High-Density): Configured for high-concurrency inference, video analytics, and cloud gaming with support for up to 16 single-width FHFL or 20 HHHL PCIe GPUs.

Key Features & Highlights

• One-Click Software Topology Switching: The industry’s first server to support software-level switching between Balance, Common, and Cascade topologies on the NF5468M6-P model to adapt to varied deep learning communication patterns.

• PCIe Gen4 Performance: Powered by PCIe 4.0 architecture, doubling I/O bandwidth compared to PCIe 3.0 to sustain demanding GPU-to-CPU and GPU-to-NVMe transfer requirements.

• Broad Accelerator Ecosystem: Compatible with NVIDIA (A100, A40, A30, A10), AMD (Instinct MI100), Intel, and domestic AI accelerator cards.

• Massive Local Storage: Configurable with up to 24x 2.5″ or 12x 3.5″ drives, including support for up to 8 hot-swappable NVMe SSDs and onboard dual M.2 boot drives.

• High Redundancy & Thermal Efficiency: Features up to 4x hot-swappable Platinum PSUs (2+2 / N+N redundancy up to 3000W each) and N+1 redundant cooling fans.

Technical Specifications