04 / Services
Liquid Cooling Integration_
CDU installation, rack-level manifold routing, quick-disconnect fittings, leak detection, and thermal commissioning for liquid-cooled GPU platforms. GB200 NVL72 and GB300 NVL72 racks require liquid cooling — there is no air-cooled option. We integrate it.
Required for all NVIDIA GB200 and GB300 NVL72 deployments
Why Liquid Cooling Is Required_
The GB200 NVL72 rack draws approximately 120kW. The GB300 NVL72 draws more. At these power densities, air cooling physically cannot remove heat fast enough. The thermal design power of 72 Blackwell GPUs in a single rack exceeds what fans and heat sinks can dissipate.
Direct liquid cooling circulates coolant through cold plates mounted on GPU and CPU packages, removing heat at the source. It is not an upgrade or an optimization — it is the only viable thermal solution for current-generation high-density GPU platforms.
Every GB200 and GB300 NVL72 deployment includes liquid cooling as a core infrastructure requirement, not an add-on. Our scope covers the full cooling integration from CDU to cold plate.
Scope of Work_
CDU Installation
Coolant Distribution Unit placement, connection to the facility chilled water loop, and commissioning. We install, connect, and validate CDU operation including heat exchanger performance, pump function, filtration, and control system configuration.
Rack-Level Manifold Routing
Supply and return manifolds routed from CDU to each GPU rack. Manifolds distribute coolant evenly across all server trays. Proper manifold design ensures balanced flow — every GPU receives adequate cooling.
Quick-Disconnect Fittings
QD fittings installed at each server tray allow individual servers to be removed for maintenance without draining the entire cooling loop. We install QD fittings, verify seal integrity, and validate torque specifications.
Leak Detection
Leak detection sensors installed under each rack, at CDU connections, at manifold junctions, and at quick-disconnect points. Sensors wired to monitoring systems that generate alerts before coolant reaches electrical components.
Pressure Testing & Flushing
The entire cooling loop is pressure tested before coolant is introduced. This identifies leaks, weak fittings, or manufacturing defects. After pressure testing, the loop is flushed to remove particulates.
Coolant Fill & Flow Balancing
Coolant introduced to the loop with controlled fill procedures to avoid air pockets. Flow rates balanced across all racks and server trays to ensure uniform cooling.
Thermal Commissioning
GPU racks brought under load and thermal performance validated. Coolant inlet and outlet temperatures, flow rates, and GPU junction temperatures monitored to confirm the cooling system meets platform requirements.
CDU Architecture_
A CDU (Coolant Distribution Unit) contains a heat exchanger, circulation pumps, filtration, a reservoir, and a control system. Facility chilled water (typically 10–20°C) flows through the primary side of the heat exchanger. Rack coolant flows through the secondary side.
The CDU maintains coolant at the target temperature and flow rate regardless of GPU load fluctuations.
One CDU typically serves one to two GPU racks depending on the platform and thermal load. GB200 NVL72 deployments at 120kW per rack often require dedicated CDUs per rack.
The CDU must be sized to handle peak thermal load with margin — undersized cooling is a deployment failure that cannot be fixed without hardware changes.
Coordination With Facility Mechanical_
Liquid-cooled GPU deployments require coordination between the IT infrastructure team and the facility mechanical team. The building must provide chilled water at the correct temperature, flow rate, and pressure to feed the CDUs. Piping, valves, and isolation points must be in place before GPU rack installation begins.
We coordinate with facility mechanical teams and general contractors to ensure the building-side water infrastructure is ready for GPU cooling integration. This includes reviewing mechanical drawings, confirming flow capacity, and validating water quality before connecting CDUs.
Track Record_
Active GB200 NVL72 and GB300 NVL72 liquid-cooled deployments. CDU installation, manifold routing, leak detection, and thermal commissioning across hyperscale AI training facilities.
Our technicians have hands-on liquid cooling deployment experience at facilities operated by Meta, Oracle, xAI, and Computacenter.
Related Resources
Ready to Deploy Your GPU Infrastructure?_
Tell us about your project. Book a call and we’ll discuss scope, timeline, and the best approach for your deployment.
Book a Call