What Is Two-Phase Cooling?_
Two-phase cooling removes heat by letting a coolant boil — change from liquid to vapor — at the heat source, exploiting the large latent heat of vaporization to absorb heat at near-constant temperature. It appears in two forms relevant to GPUs: two-phase immersion, where servers sit in a low-boiling dielectric fluid, and two-phase direct-to-chip, where coolant boils inside cold plates. It contrasts with single-phase cooling, where the coolant stays liquid and only warms.
Technical Details
In single-phase liquid cooling, coolant absorbs heat by rising in temperature, so heat capacity is bounded by flow rate and allowable temperature rise. Two-phase cooling instead absorbs heat through the phase change from liquid to vapor, which carries far more heat per unit mass and holds the surface near the fluid's boiling point — beneficial for keeping high-power dies within tight temperature limits. Two-phase immersion uses engineered dielectric fluids that boil at low temperature; vapor rises to a condenser and returns as liquid. Two-phase direct-to-chip boils coolant inside the cold plate. The trade-offs are fluid cost and handling, condenser and containment design, and the maturity of supporting infrastructure. Current NVIDIA NVL72 systems use single-phase direct-to-chip cooling rather than two-phase, though two-phase is an active area of development for higher densities.
How Leviathan Systems Works with Two-Phase Cooling
Leviathan Systems primarily deploys single-phase direct-to-chip cooling for current NVIDIA platforms and advises on facility design considerations where two-phase cooling is under evaluation for future densities.
Related service
Liquid Cooling Services →