Retrofitting 20 kW Racks into a 5 kW Design Basis Hall
An operator wants to add 320 kW of high-density compute to a hall whose design basis was 5 kW per rack. The hall has 12 rows of eight racks, a 450 mm raised floor, 480 kW of installed CRAH capacity, and under-floor power whips from panelboards at the room perimeter. The proposal is two new rows of eight racks at 20 kW each, filled with blade chassis, plus a busway run overhead. The structural drawings state a live load capacity of 8 kN/m² for the computer floor and 6 kN/m² for the general slab; the raised floor is 600 × 600 mm tiles on 600 mm pedestals. Current IT load is 380 kW and the measured PUE last month was 1.9. The business case assumes the existing CRAHs will 'absorb' the extra load because the room is only at 79% of design kW. Facilities has been asked only to confirm that power capacity exists at the perimeter panelboards.
1. What must be verified before this can be approved, and what do the numbers already tell you?
Read the worked answer
Perimeter power is the least of the issues. Four of the constraints can already be assessed from the numbers given.
- Cooling capacity fails outright. 380 + 320 = 700 kW of IT load against 480 kW of installed CRAH capacity - 146%. The '79% of design kW' statement refers to current load versus installed cooling, and it is being used to justify load that the plant cannot remove. Whatever else is true, the cooling must be upgraded or the load cut.
- Floor loading must be proven, not assumed. A fully populated blade rack with chassis, in-rack PDUs and cabling commonly masses 1.0–1.4 t. On a 600 × 1200 mm footprint (0.72 m²) that is roughly 14–19 kN/m² locally, against a stated 8 kN/m² floor live load and 6 kN/m² slab. A raised floor concentrates load onto pedestals, so the structural check is per-pedestal point load plus the average, not average alone. No approval without the structural engineer's written assessment.
- Airflow cannot be delivered. 20 kW per rack needs on the order of 5,000 m³/h at a 12 K rise (P = 1.2 × volumetric flow × ΔT), so about 80,000 m³/h for 16 racks. A 450 mm plenum limits achievable static pressure, and standard tiles pass roughly 800–1,200 m³/h each - meaning five or more high-flow grates per rack, or ducted/direct supply.
- Distribution is wrong for the density. 20 kW at 400 V three-phase is about 29 A per phase. Perimeter whips give long runs, voltage drop, and no A/B path. The busway instinct is right, but it must be dual-path for dual-corded IT.
And a signal about the plant: a measured PUE of 1.9 says cooling and distribution are already inefficient, so adding density to the same infrastructure will make it worse.
2. Which standards and design references govern this decision, and what specifically do they drive?
Read the worked answer
Several references apply, each to a different question - citing one alone invites a wrong answer.
- Data centre topology standards (TIA-942, EN 50600 series, ISO/IEC 22237): drive the redundancy and path-separation requirements for the new rows - whether the busway is single or dual path, whether the new cooling has N+1, and how the new load is separated from the existing installation. They also drive the documentation set the change should produce.
- ASHRAE thermal guidelines: set the inlet temperature and humidity envelope that the cooling design must hold at 20 kW per rack, and therefore whether containment and rack-level cooling are optional or mandatory.
- Electrical installation code (IEC 60364 or NFPA 70 as applicable locally): governs conductor sizing, voltage drop, overcurrent protection, and selective coordination between the busway tap-offs and the upstream breakers - a coordination study is required before energisation, not after.
- Raised access floor standard (for example EN 12825 or the equivalent regional standard) and the structural engineer's assessment against the building code: define the tile and pedestal load ratings and whether point loads must be spread with plates or a frame.
- Fire protection and life safety codes: higher power density changes the fire load and the design density for sprinkler or clean-agent protection, and egress and compartmentation must be re-checked.
The approval should be conditioned on all five being satisfied for the specific new rows, not on a general statement that the room meets the standard.
3. If the density must stay at 20 kW per rack, describe a design that would actually work and the phasing you would insist on.
Read the worked answer
A workable design for these two rows would look like this:
- Cooling: contained cold or hot aisle for the new rows with in-row or rear-door heat exchangers, because 20 kW per rack cannot be reliably served by perimeter CRAHs on a 450 mm plenum. Supply the rear-door units from the existing chilled water loop only after confirming the loop has the capacity and the pumping head for the extra coils.
- Airflow: high-flow grates or direct ducted supply inside the containment, with commissioning airflow measurements per rack rather than per room.
- Power: dual overhead busway with A and B sources, tap-off boxes per rack, and 2 × three-phase rack PDUs in each rack so every dual-corded device has two independent paths. Verify selective coordination to the upstream breakers.
- Structure: spreader plates or a slab-mounted frame under the new rows, sized to the structural engineer's per-pedestal allowance, with the rack layout drawn to show pedestal positions.
- Cooling plant: add CRAH or chiller capacity to cover the full 700 kW, and treat the PUE of 1.9 as a separate remediation project - it will otherwise erode the business case for the new compute.
Phasing, which I would insist on: build and commission two racks only as a pilot, with containment, busway and rear-door cooling in place, and measure inlet temperatures, airflow and actual rack power against design for a month. Then release the remaining fourteen racks. This is a change to the design basis of the hall, not a fit-out; the pilot is what converts the engineering assumptions above into verified numbers, and it is far cheaper to discover a pedestal or airflow problem with two racks than sixteen.