AI/HPC Liquid Pod
Choose a cooling family, expose every remaining kW of air load, stress the liquid train and add project cost or water assumptions only when you have them.
This interactive teaching scenario begins with a 900 kW, twelve-rack AI/HPC pod and a direct-to-chip starting assumption. Those values are editable scenario inputs, not typical performance claims. Air-only, rear-door heat exchanger, direct-to-chip and immersion buttons load different starting assumptions for heat capture and available path capacity; they do not rank products or declare a universal best cooling family.
Learning objectives
- Trace heat from silicon through technology cooling liquid, CDU and facility rejection
- Calculate the flow implied by heat load, fluid properties and coolant delta-T
- Test liquid-train and residual-air failures without declaring a universal best technology
One heat balance, two possible paths
Every entered kilowatt of IT load is assigned to handled liquid heat, handled residual-air heat or an explicit thermal gap. The liquid route follows silicon to a technology loop, CDU or heat exchanger, facility-water loop and ambient rejection or coincident useful-heat demand. The air route preserves the heat not captured by liquid equipment and compares it with the entered residual-air capacity. The model never treats a liquid-cooling label as proof that room cooling can disappear.
A text table repeats every diagram connection with source, destination, carrier, heat quantity and current state. The same pure evaluator drives the graphic, constraint register and automated heat-conservation tests.
Flow and temperature boundary
Calculated coolant flow uses the handled liquid heat, entered fluid density, entered specific heat and entered coolant temperature rise. Technology-coolant supply, facility-water supply and minimum heat-exchanger approach remain separate assumptions. A fouling drill increases the required approach, reduces available liquid capacity and raises the modeled liquid-loop power rather than hiding degradation inside a generic efficiency score.
The calculation does not size pipes, valves, pumps or heat exchangers. It omits pressure drop, hydraulic balancing, transient behavior, water chemistry, controls, vendor performance curves and commissioning tolerances.
Failure drill
Select pump loss, valve isolation, leak alarm, fouling or residual-air loss. A single modeled train cannot survive its pump loss; redundant train assumptions can retain one available train. Valve isolation and the illustrative leak-alarm sequence remove the selected pod liquid branch. Residual-air loss exposes exactly how much heat the selected liquid capture did not remove. The outcome reports stable, degraded or thermal-gap state without inventing an uptime percentage.
Energy, cost, water and heat reuse
PUE, facility energy, cooling support and residual-air heat are compared with a notional air-only path under the same IT load and residual-air COP. If any IT heat lacks a modeled cooling path, the lab withholds PUE, energy-cost and water comparisons: lower cooling power during a failed state is not presented as an efficiency improvement. Installed cost, annual energy cost and annual water otherwise remain “not evaluated” until the learner enters project-specific cost range, tariff and facility water-intensity assumptions. Zero deliberately means no claim. Useful-heat output is limited by both handled liquid heat and coincident entered demand; temperature quality and delivery infrastructure remain unresolved.
Use and limitations
Use this model to understand boundaries, form better questions and record a reproducible assumption set. It is not measured site data, a vendor comparison, a construction design, a code-compliance review or a commissioning result. Real work requires verified loads, equipment data, fluid compatibility, structural and electrical coordination, fire/life-safety review, site engineering and current authority requirements.
Save and compare versions in Design Pro or open the hybrid liquid-zone reference card.