What Is Direct-to-Chip Cooling?_
Direct-to-chip cooling (also called cold-plate liquid cooling) routes liquid coolant through cold plates mounted directly on heat-generating dies — GPUs, CPUs, and sometimes memory or voltage regulators — capturing heat at its source rather than relying on chassis airflow. It is the cooling method used by NVIDIA NVL72 systems and is the dominant approach for high-density GPU racks. Direct-to-chip is a form of direct liquid cooling and is distinct from immersion cooling, where the entire server is submerged.
Technical Details
In a direct-to-chip system, coolant — typically a treated propylene glycol and water mixture — flows from a manifold through flexible hoses and quick-disconnect fittings into cold plates seated on each die via thermal interface material. The warmed coolant returns to a coolant distribution unit (CDU), which transfers heat across a heat exchanger to the facility water loop. Direct-to-chip cooling commonly captures 70–90% of rack heat at the chip, with any residual handled by air or a rear-door heat exchanger; some designs add cold plates to additional components to push the liquid-captured fraction higher. The approach scales to the roughly 120 kW per-rack loads of NVL72 systems, where air cooling is not viable. Deployment quality hinges on correct cold-plate mounting pressure, leak-tight fittings, manifold balancing, and leak-detection coverage.
How Leviathan Systems Works with Direct-to-Chip Cooling
Leviathan Systems integrates direct-to-chip cooling for high-density GPU deployments, connecting cold plates and manifolds, installing leak detection, and pressure-testing the loop during commissioning.
Related service
Liquid Cooling Services →