Rack heat load and coolant flow calculator
Calculate required coolant mass flow rate and volumetric flow from rack heat load, temperature rise, and fluid properties. Based on the fundamental energy balance Q = m × cp × ΔT.
How we calculated this
The calculator uses the fundamental energy balance equation: Q = m × cp × ΔT, where Q is heat load (W), m is mass flow rate (kg/s), cp is specific heat capacity (J/kg·K), and ΔT is temperature rise (K).
Fluid properties are approximate values at ~25°C. Specific heats are from Engineering ToolBox water-glycol tables (% by mass): propylene glycol 30%/40% = 0.935/0.895 Btu/lb·°F (https://www.engineeringtoolbox.com/propylene-glycol-d_363.html); ethylene glycol 30%/40% = 0.912/0.874 Btu/lb·°F interpolated between the 20°C and 30°C columns (https://www.engineeringtoolbox.com/ethylene-glycol-d_146.html); converted at 1 Btu/lb·°F = 4.1868 kJ/kg·K. Density values are approximate. Actual properties vary with temperature and concentration.
The safety margin (default 1.3×) accounts for GPU power transients, manufacturing tolerances, and future density increases. NVIDIA GB200 NVL72 racks approach 140 kW (TrendForce, 2024; NVIDIA, 2026).
ASHRAE TC 9.9 liquid cooling classes (W17-W45) define facility water supply temperature envelopes (2°C lower limit up to the named maximum). The ΔT across cold plates comes from the IT equipment spec; 10°C is a typical direct-to-chip design value.
The result is the flow required at the rack manifold. CDU pump must overcome additional pressure drop from cold plates (5-20 psi each), manifolds, QDs, piping, filtration, and heat exchanger.
Method checked on 2026-11-01.
Engineering estimate only. Not a substitute for vendor-qualified thermal design. Confirm all values with your CDU supplier, cold-plate manufacturer, and facility engineer.
Questions
- Is this a capacity guarantee?
- No. This calculator provides engineering estimates based on the formula Q = m × cp × ΔT. Confirm actual flow requirements with your CDU vendor and cold-plate manufacturer.
- What safety margin should I use?
- Typical practice is 1.2-1.5× the nominal heat load to account for transient GPU power spikes and future density increases. NVIDIA GB200 NVL72 racks approach 140 kW (TrendForce, 2024).
Find vendors
Use this tool on your site
The snippet includes a link back to this page. Please keep it.