The Dell PowerEdge XE8640 is a high-performance 4U rack server specifically engineered for AI, HPC, and analytics workloads. It offers an optimal balance of compute power and GPU acceleration, featuring a unique cooling solution that allows for high-wattage components in a standard air-cooled environment.

Full Technical Specifications

Component	Specification
Form Factor	4U Rack Server
Cooling	Liquid Assisted Air Cooling (LAAC) for GPUs; Air Cooling for CPUs and system. (Note: No facility water to rack required.)
Processor	Dual 4th or 5th Generation Intel® Xeon® Scalable Processors (up to 64 cores per processor)
Accelerators (GPU)	4x NVIDIA HGX H100 80GB SXM5 GPUs, fully interconnected with NVIDIA NVLink technology
Memory	32 DDR5 DIMM slots, max 4 TB RDIMM, speeds up to 5600 MT/s (with 5th Gen Intel)
PCIe Expansion	Up to 4 PCIe Gen 5 slots (4 x16) for high-speed I/O
Storage (Front Bays)	Up to 8 x 2.5-inch NVMe SSDs OR up to 8 x E3.S NVMe SSDs, max 122.88 TB
Boot Storage	Boot Optimized Storage Subsystem (NVMe BOSS-N1) with HWRAID 1 (2 x M.2 SSDs)
Power Supplies	2800W / 3200W Titanium (Hot-swap redundant)
Networking	1 x OCP 3.0 (optional), 2 x 1GbE LOM
Management	iDRAC9 with GPU telemetry, OpenManage Enterprise (OME)
Workloads	AI/ML/DL Training (medium to large datasets), HPC Modeling & Simulation, Data Analytics

Description

The Dell PowerEdge XE8640 is a high-performance, 4U rack server designed to drive demanding AI, HPC, and analytics workloads using a powerful 4-way GPU architecture. It is built around dual Intel Xeon Scalable processors and features four NVIDIA HGX H100 SXM5 GPUs that are fully interconnected via NVIDIA NVLink for maximum GPU-to-GPU bandwidth (⁵ $\approx 900 \text{ GB/s}$ ).

The server's unique advantage is its Liquid Assisted Air Cooling (LAAC) system. This hybrid approach uses liquid cooling plates to manage the intense heat of the GPUs while relying on high-performance fans for the rest of the components. This design allows the XE8640 to deliver the superior performance of an HGX platform and high-wattage components without requiring a facility liquid cooling infrastructure, making it simpler to deploy in standard air-cooled data centers up to $35^\circ \text{C}$ ambient temperature. This makes it an ideal choice for businesses looking to accelerate AI training and simulation with a fast time-to-value.