Leak-detection calculator

Companion tool · Leak-detection deep-dive

Leak-detection topology model

An author-built engineering scenario model. Every count, cable metre, controller, endpoint and cost figure is derived from four inputs and adjustable assumptions. Not a vendor design guide or industry-standard bill of materials.

4 inputs · reference 100 MW hall · sensor + coverage topology model · free tool
100 MW gross → 85 MW IT → 945 racks → 79 pods → 99 CDUs → 20.8 km cable → 1,445 sensing + 99 telemetry → $1.25M

Inputs

Total campus power draw at design load. IT MW derives via load factor.
IT MW / gross MW. Reflects PUE and non-IT allocations.
NVL72-class Rubin racks land 60-120 kW. Older H100 halls 30-60 kW.
Row-based is the reference architecture for this model. Select in-rack when modelling a 1:1 CDU architecture.
Reference assumption: 1.25 = one additional CDU per four pods (illustrative N+1 allocation). 1.00 = N-only, 1.50 = one extra CDU per two pods, 2.00 = full redundant pairs. Row-based mode only. Scenario choice, not an industry standard.
Tier 1 strips pod/rack coverage. Tier 3 adds redundant pod hubs plus optional slot-level ribbon.
Metres of sensing cable per metre of hall wall. Includes multiple parallel runs along perimeter, plant-room walls and cold-aisle boundaries. Hall wall length derives from area per rack and aspect ratio (Advanced).
Greenfield sets install ~25%. Retrofit sets install ~50% (cable pulls in a live hall are expensive). Reference is 35%. Adjust slider below to fine-tune.
Field integration, cable pulls, commissioning as a percentage of hardware. Highly site-dependent.
Optional. For immersion or two-phase environments where resistive cable cannot see the coolant. Not required for conventional single-phase DTC.
Advanced · geometry and unit costs
Gross area per rack is the allocated facility footprint including plant, aisles and non-white-space, not white-space density. Perimeter cable = 2 × (L + W) × cable density, where L and W come from area × aspect ratio.
Illustrative installed detection cost
$1.26M
$14.8/kW IT · $1,300/rack
Labour band (install 30% → 50% of hardware): $14.2 to $16.4/kW IT
Illustrative reference topology. Not a vendor quotation or design specification.
85 MW
IT capacity
945
Racks
79
Pods
99
CDUs

Coverage density

23.1
Cable m per rack
257
Cable m per MW IT
1.53
Physical detection points / rack
1.63
Total monitored signals / rack

Cable distribution

Facility / perimeter3,820 m
Row / pod14,220 m
Rack3,780 m
Cable metres are topology assumptions rather than vendor specifications.

Detection topology

LayerCable mSpotsControllersCDU telemetry
Facility / perimeter3,8206512—
Row / pod14,22023779—
Rack3,780945via pod—
CDU (physical + telemetry)—198via pod99
Total21,8201,4459199
The Spots column counts physical detection points (rope-cable segments, pucks, dielectric probes and CDU-internal spots). The CDU-telemetry column counts indirect signals off the CDU (reservoir level, pressure, flow, alarm channels) that support leak detection through pressure or flow anomaly. Physical detection points (1,445) and indirect telemetry endpoints (99) together comprise 1,544 monitored signals in the reference case.

Cost breakdown

LineCount$/eachSubtotal
Sensing cable21,820 m$8$175k
Physical detection points1,445$180$260k
Pod controllers / hubs79$3,500$277k
Master locator panels12$8,000$96k
CDU telemetry endpoints99$400$40k
Integration gateways14$6,000$84k
Dielectric overlayoff——
Hardware subtotal——$931k
Installation + field integrationat 35% of hardware$326k
Illustrative installed cost——$1.26M

Detection ≠ localisation ≠ diagnosis ≠ response

Detection
Did liquid escape?

Sensing cable, spot sensors, dielectric probes. Any layer can answer this. The bar is low.

Localisation
Where did it escape?

Addressable cable, pod-level controllers, master locator panels. Only some architectures provide this.

Diagnosis
Which subsystem caused it?

Combining spot, cable, and CDU telemetry to distinguish a QD leak from a manifold pinhole. Software layer, not hardware.

Response
What action follows?

Hardwired dry-contact to CDU solenoid, BMS alarm, isolation playbook. The value migrates here.

Four-scenario sensitivity

ScenarioRack densityTierRacksCable mEndpoints$/kW ITInstalled cost
Minimum60 kW11,4174,605363$6.2$528k
Reference90 kW294521,8201,544$14.8$1.26M
High density120 kW270916,9871,179$11.2$954k
High coverage90 kW394526,8701,623$20.0$1.70M
Same 100 MW campus, four topology assumptions. Watch how far the headline number moves between Minimum and High-coverage tiers. Topology is the driver.
This is an author-built topology model. It is not a vendor design guide or industry-standard bill of materials.
Illustrative only. Default values are author assumptions intended to show how coverage and cost scale with topology. Change them to test your own architecture. Outputs are scenario estimates rather than procurement quotations.

Method

Everything derives from four primary inputs: gross MW, IT load factor, kW/rack, racks/pod. Everything else is either a coverage assumption you can override, or a unit cost you can override.

Rack count = ceil(IT MW × 1000 / kW per rack). Pod count = ceil(rack count / racks per pod). CDU count in row-based mode = ceil(pod count × CDUs per pod); in-rack mode = rack count.

Hall footprint is derived from gross facility area per rack (default 90 m², editable in Advanced) times rack count. Total footprint is split across multiple halls above 50 MW IT (one hall per 50 MW IT), so large campuses aggregate wall length across buildings rather than model an implausibly large single rectangle. Each hall is a rectangle with aspect ratio L:W (default 2:1, editable). Perimeter per hall = 2(L + W) where W = √(area / aspect) and L = aspect × W. Facility cable = aggregate perimeter × cable density (default 2.0 m of sensing cable per m of wall, includes multiple parallel runs along perimeter, plant walls and cold-aisle boundaries) + chilled-water riser runs.

Pod cable = 180 m × pod count in Tier 2 (0 m in Tier 1, 220 m in Tier 3). Rack cable = 4 m × rack count in Tier 2 (0 in Tier 1, 6 m in Tier 3).

Tier 3 adds a pod hub redundancy multiplier and optional dielectric overlay. Tier 1 strips pod and rack cable and rack spots entirely (perimeter and CDU only). Integration gateways scale at approximately 1 per 6 pods. Master locator panels scale at approximately 1 per 7 pods.

Unit costs are hardware only. The installation + integration factor is applied on top and is highly site-dependent (new build vs retrofit, cable routing, commissioning, contractor labour, geography).

Every value below the four primary inputs is an editable illustrative assumption. Change any of them to model your own scenario. Source discussion and named vendors in the leak-detection deep-dive.

← Back to the leak-detection deep-dive
Sibling calculator · parent-essay economics

DTC cooling · capex, opex + TCO calculator

If this leak-detection tool is telling you what the sensor and control layer costs, the DTC cooling calculator tells you what the underlying cooling architecture costs. Six architectural inputs, capex waterfall from the Fig C15 anchor, year-1 opex, 7-year TCO, payback engine, sensitivity matrix and scenario-sharing links. Same reader path.

Underlying architecture deep-dive

Data Centre Cooling · a double-click on Direct-to-Chip

The parent architecture essay for both this leak-detection deep-dive and the DTC calculator. Nine-layer physical stack from GPU die and TIM through cold plate, QDs, manifold and CDU out to the facility loop. CDU as testable control-point proposition. Consolidation race (Ecolab-CoolIT, Schneider-Motivair, Eaton-Boyd, Vertiv-STL/PurgeRite). Investor decision map by profile.