Products / Cold Plate

Chip-level liquid-cooling cold plate

A chip-level direct-liquid cold plate for GPUs, CPUs and AI accelerators. Microchannel, jet-impingement or hybrid flow paths can be developed around the device heat map, mechanical envelope, flow and pressure-drop budget, with evidence built through simulation, prototypes and liquid-loop testing.

550 W

Measured total heat load

300 W/cm²

Average heat flux (local 608)

91 °C

Chip peak temperature (limit 120 °C)

5.8 kPa

Pressure drop @1.5 L/min

Engineering context

Understand the system before selecting the component

AI, HPC and high-density servers concentrate heat around processors, accelerators and memory interfaces. An air-only path can make local heat flux, airflow, noise and rack power competing constraints.

Principle

How this product works in the cooling path

A cold plate couples to the package through a thermal interface and directs coolant through controlled channels close to the heat source. The warmed coolant then transfers heat through the CDU or facility loop.

From heat source to facility loop

Step 1: Confirm package, heat-source and mechanical boundaries
Step 2: Establish the thermal path through the interface material
Step 3: Design channels around flow, pressure-drop and temperature windows
Step 4: Validate heat rejection, monitoring, service and protection at system level

What the design helps you control

Captures heat at the source for high-heat-flux devices
Coordinates cold plates, fittings, tubing and controls as one interface
Can operate with air cooling, rear-door heat exchangers or facility-water systems

Inputs and application boundaries

Package, interface material and mounting load must be defined
Coolant, flow, pressure, temperature and water-quality requirements must be aligned
Uncovered memory, power and auxiliary devices may still require residual air cooling

Typical engineering path

Heat-source and workload profile
Prototype and thermo-fluid validation
Server or rack integration
Commissioning, monitoring and service handover

Engineering details

What must be confirmed before release

Role in the cooling system

The cold plate sits between the chip package and the server secondary loop, moving heat into the coolant at the source before connecting downstream to server tubing, quick connects, the rack manifold and the CDU.

Inputs required for selection

Chip geometry, package and mounting load, plus steady, peak and spatial heat maps
Allowable junction temperature, surface uniformity, coolant and supply-temperature range
Available branch flow, pressure-drop budget, pump and rack-loop conditions
Server envelope, port orientation, assembly sequence and service method

Interfaces and engineering boundaries

TIM, flatness and mounting load jointly determine package-to-cold-plate contact resistance.
Cold-plate materials, joining route, seals and coolant require compatibility review.
Cold-plate pressure drop must be included in the server branch, manifold and CDU system-resistance budget.

Validation and acceptance evidence

Record chip temperature, cold-plate uniformity and thermal resistance under the target heat map and supply conditions
Record the flow-pressure curve, distribution and thermal response under abnormal conditions
Complete pressure integrity, leak, cleanliness and critical-dimension checks

Installation, operation and service conditions

Production release values are governed by the controlled drawing, test specification and approval record for the applicable model.

Key capabilities

Heat-map-led flow paths prioritize hotspots and temperature uniformity

Thermal resistance, pressure drop, pumping demand and server space are balanced together

Materials, precision machining, joining, sealing and leak risk are designed as one system

CFD, prototype builds and liquid-loop testing progressively correlate the design

Applications

AI server GPUs, CPUs and accelerators

Technical specifications

总热负荷550WVerified

平均热流密度300W/cm²Verified

局部峰值热流密度608W/cm²Verified

冷却液流量1.5L/minVerified

FAQ

Can the cold plate fit my chip?

Cold plates are custom-designed per package. Share package size, power and heat-flux distribution, and we design the channels and validate with CFD and prototypes.

Is 1.5 L/min enough?

Measured: 550 W load at 1.5 L/min keeps the chip at 91 °C. The exact requirement depends on load and allowed temperature rise.

What do I gain versus air cooling?

Liquid cooling captures heat at the source, eliminating hot spots and throttling while enabling higher compute density, with much lower rack noise and airflow.

What does 608 W/cm² peak mean?

It is the measured local hot-spot heat flux. The hybrid channels are designed around the heat map so local peaks do not become temperature bottlenecks.

Discuss your thermal requirements

Tell us your heat source, duty and constraints — we respond with a design and validation plan within 1-2 business days.

Talk to an engineer