OCM Modular Server Design: 2026 Standard Decouples CPU, Power, BMC for Upgradeable Infrastructure

System Overview

The Open Compute Standard Working Committee's 2026 guide 《基于OCM的模块化服务器设计指南》 defines a modular server architecture called OCM (Open Compute Module). Traditional monolithic servers require a complete motherboard and system redesign for every CPU generation. OCM solves this by separating power delivery, BMC management, and cooling into independent modules that stay in the chassis, while only the CPU-and-memory-carrying OCM motherboard is replaced during a processor upgrade.

Hardware Specifications (2U Single-Socket 19" Rack)

Compute

One OCM motherboard supporting Intel Birch Stream or AMD Turin single-socket processors.

Up to 12 DDR5 RDIMM/MRDIMM slots.

I/O Expansion

Maximum 10 PCIe add-in cards .

Supports OCP NIC 3.0 SFF hot-pluggable network adapters.

Storage Configuration

Front bay: 24 × 2.5" or 12 × 3.5" drives (SATA/SAS/NVMe, hot-plug).

Rear bay: 4 × 2.5" / 4 × 3.5" drives plus 2 internal M.2 slots.

Physical Chassis

Standard 2U height (87 mm), EIA-310-E compliant, 850 mm mounting depth. The interior is divided into interchangeable functional zones; modules in the same zone can be swapped without modifying the chassis base .

Front zone: drive modules only.

Rear zones A–F: each zone accepts PCIe expansion modules, GPU modules, drive modules, or NIC modules. For example, the same slot can take a PCIe FHHL card or a drive backplane.

System Topology

The server comprises five primary modules: OCM motherboard, power board, BMC management board, fan board, and front/rear I/O/storage modules .

Physical Layout Details

Chassis width 482.6 mm, depth 849.7 mm. Key layout points:

OCM motherboard (including CPU heatsink) mounts directly; PSU connects straight to the motherboard.

Fan modules plus 2 × SATA/PCIe M.2 drive modules are integrated.

Front bay: 24 × 2.5" or 12 × 3.5" drive modules.

Rear bay options: 3 × PCIe FHHL, 2 × PCIe FHHL, 4 × GPU, 4 × 3.5" drives, 4 × 2.5" drives, or 1 × OCP NIC 3.0 SFF/TSFF.

Front-Bay Modules

24 × 2.5": 24 bays, SAS/SATA/NVMe, hot-plug.

12 × 3.5": 12 bays (2.5" compatible), SAS/SATA/NVMe, hot-plug.

Rear-Bay Modules (Zones A–F)

Zones A and B each support either:

3 × PCIe FHHL: PCIe ×8 or ×16; accommodates three single-width or one dual-width plus one single-width card.

2 × 3.5" drives: 3.5" (2.5" compatible), SAS/SATA/NVMe, hot-plug.

Motherboard Unit

The OCM motherboard is partitioned into five regions: CPU, DIMM, high-speed connector, management connector, power connector . The management connector uses a UDIMM-form-factor connector placed beside the memory connectors.

Power Delivery

A dedicated power board feeds the OCM motherboard, fan modules, drive modules, PCIe devices, GPUs, smart NICs, OCP 3.0 NICs, and M.2 drives. External power connectors are derated; on-board loads and connector front-ends include power monitoring and fuses .

Interconnect

High-speed: MCIO ×8 connectors .

Low-speed (fan board, ear board, power board, intrusion detection, TPM, PCIe riser, drive backplane, leak detection): all comply with OCM standard pinouts.

Management Module

The BMC chip moves onto a separate management board that communicates with the motherboard via a defined interface. This decoupled module can be designed and upgraded independently of the main board.

Fan Module

Independent fan module with six 6056 fans (other form-factors swappable).

Hot-plug, N+1 redundancy , tool-less install/remove, anti-misinsertion keying.

Fan board is physically separate from the motherboard; powered via the power board; interconnect follows OCM standard.

CPU Air Cooler

Two-piece design for 2U single-socket 550 W TDP:

Main heatsink: 126.8 × 100 × 64 mm, aluminum, 63 fins (0.3 mm thick), 7 × Φ6 heat pipes.

Extension heatsink: 304 × 70 × 47 mm, aluminum, 191 fins (0.3 mm thick).

Future Outlook

The guide aims to accelerate OCM standard adoption across the supply chain: validating feasibility with reference designs, unifying terminology, and shifting the industry from fragmented innovation to standardized collaboration—thereby lowering integration costs and sustaining long-term server innovation.