Comprehensive Liquid‑Cooling Reference Design for Server Components (2024)

The document introduces a full‑liquid‑cooling reference design (2024) that extends beyond the commonly explored CPU and GPU cooling to high‑power memory, SSDs, OCP NICs, PSU, PCIe and optical modules, achieving the industry's highest cooling coverage for diverse deployment needs.

1. CPU Cold‑Plate Design – A reference CPU cold‑plate based on Intel 5th‑gen Xeon platforms, optimized for thermal performance, structural integrity, yield, cost and material compatibility, comprising an aluminum bracket, cold‑plate and connectors.

2. Memory Liquid‑Cooling Design – An innovative “sleeper‑rail” heat‑sink solution combining traditional air‑cooling and cold‑plate methods, using built‑in heat pipes or vapor chambers to transfer heat to the cold‑plate via thermal pads, offering easy maintenance, broad compatibility, high cost‑effectiveness, manufacturability and reliability.

3. SSD Liquid‑Cooling Design – A liquid‑cooled SSD module with an integrated heat‑pipe heat‑sink, thermal‑pad contact to a dedicated cold‑plate, and a drawer‑style bracket for compact installation, supporting >30 hot‑swap cycles, low‑cost manufacturing and multi‑drive sharing of a single cold‑plate.

4. PCIe/OCP Card Liquid‑Cooling Design – PCIe cards are equipped with a custom liquid‑cooled heat‑sink that interfaces with system cold‑plates; OCP 3.0 cards use a similar approach with spring‑screw locking mechanisms to ensure reliable long‑term contact and easy maintenance.

5. IO Cold‑Plate Solution – A multifunctional IO cold‑plate that cools motherboard I/O components as well as PCIe and OCP cards, built from an aluminum alloy body with copper fluid channels, optimized for layout and thermal‑interface material compatibility.

6. Power‑Supply Cold‑Plate Design – A liquid‑cooling solution for PSUs that adds an external air‑liquid heat‑exchanger to the existing fan‑cooled power supply, reducing cabinet heat, requiring no new PSU design, and supporting both distributed and centralized heat‑exchanger architectures for rack‑level deployment.

The white‑paper also provides related links to additional reference designs, AI accelerator cooling, component specifications, and a collection of AI‑related deep‑dive reports.