Liquid Cooling — Direct-to-Chip & Rear-Door
This is the flagship mechanical course of the SAVRN Learning Institute, covering direct-to-chip (D2C) and rear-door heat-exchanger liquid cooling for high-density AI accelerator racks. Students learn cold-plate and manifold construction, coolant distribution unit (CDU) operation and redundancy, coolant chemistry and multi-stage filtration, thermal interface material (TIM) application, and leak-detection and response procedures for a technology where an undetected leak can destroy hundreds of thousands of dollars of GPU hardware in minutes. The reference platform is the NVIDIA VR200 NVL72 rack (190-230 kW per rack, no air-cooled option, 45C warm-water direct-liquid-cooling inlet) and the GB300 NVL72 rack (132-142 kW per rack); students work with the AL-2 pod's NVL72-form mechanical rack, cold-plate thermal emulator array, in-row and sidecar CDUs, and rear-door heat exchanger. Students compare dielectric single-phase fluid to water-glycol secondary-loop chemistry, perform CDU filter changes at the 25-50 micron secondary-loop specification, and execute a full leak-response drill. Every competency involving live coolant flow, a pressurized quick-disconnect, or a powered GPU tray is treated as safety-critical.
What you'll be able to do
- Diagram the direct-to-chip cooling loop from facility water system (FWS) through the CDU heat exchanger to the technology cooling system (TCS) secondary loop, cold plates, and manifold return.
- Compare dielectric single-phase immersion/cold-plate fluid to water-glycol secondary-loop chemistry, identifying the maintenance and material-compatibility tradeoffs of each.
- Connect and disconnect a pressurized blind-mate or manual quick-disconnect (QD) coolant fitting on a live cold-plate loop without introducing air or spilling coolant. Safety-critical
- Apply thermal interface material (TIM) to a cold-plate/die interface following manufacturer coverage and thickness specification.
- Operate a coolant distribution unit (CDU) in N+1 redundant configuration, including a live pump failover test. Safety-critical
- Perform a CDU secondary-loop filter change at the manufacturer-specified micron rating without introducing air or contamination into the loop.
- Detect and respond to a simulated coolant leak using the rack's integrated leak-detection sensors and the site leak-response procedure. Safety-critical
- Verify facility water supply temperature is within the ASHRAE TC 9.9 W-class band required by the deployed rack's cooling specification before commissioning coolant flow to a live rack. Safety-critical
- Operate a rear-door heat exchanger (RDHx) and its associated dry cooler, verifying supply/return delta-T and airflow are within manufacturer specification.
- Document a coolant loop commissioning, filter change, or leak-response event in the liquid-cooling maintenance log.
- Escalate a persistent leak-detection false alarm, an out-of-spec filter differential pressure, or a CDU failover event to the responsible facility engineer or manufacturer field service. Safety-critical
- Participate in a supervised live-block commissioning of a powered VR200 or GB300 NVL72-class rack under an AL-3 or employer field-placement arrangement. Safety-critical
Train the team that runs the factory.
The Institute travels with every SAVRN campus.