How should cold-plate and immersion cooling be selected?

Support / Technical answers

How should cold-plate and immersion cooling be selected?

Direct, data-backed answers to the questions that matter before and during deployment.

Cold-plate cooling moves coolant through plates coupled to chips (indirect); immersion contacts devices directly with media. Cold plate suits mainstream AI/HPC (mature, serviceable, cost-controlled — 550 W-class chips measured at 91 °C); immersion suits extreme density but costs more in media and operations.

01

The essential difference

Cold plate: TIM couples the plate to the package, coolant flows in channels (indirect), low media volume, easy service. Immersion: devices sit directly in media (single-phase) or use phase change (two-phase); shortest heat path but higher media cost and operational complexity.

02

How to choose

Mainstream GPU/CPU servers take cold plate (mature, 600 W/cm² peak validated); evaluate immersion only for extreme density or unusual geometry — start with a route assessment (media/materials/fire/O&M/pressure/rejection).

03

What to watch

Cold plates are custom per package with a residual air boundary; immersion needs media-material compatibility, fire-code and operations review. Hybrid architectures are common.