Worked Example: Data Centre UPS Output — 500 kVA UPS with Parallel Redundancy

This example demonstrates the critical but often-missed interaction between UPS output sizing and harmonic neutral current. Even with modern active-PFC power supplies that produce relatively low individual harmonic distortion, the aggregate effect of hundreds of servers creates a neutral current that exceeds the phase current — a condition that standard cable sizing tables do not address.

Each UPS operates at 48% load in N+1 configuration. During a single-UPS failure, the surviving UPS carries the full 480 kW load.

Normal operation (load sharing, 240 kVA each):

I_phase,normal = S / (√3 × V) = 240,000 / (√3 × 400) — (Eq. 1)

I_phase,normal = 240,000 / 692.8

I_phase,normal = 346 A per phase, per UPS

Contingency (single UPS, full load):

I_{phase,contingency} = 480,000 / (√3 × 400) — (Eq. 2)

I_{phase,contingency} = 480,000 / 692.8

I_{phase,contingency} = 693 A per phase, per UPS

The cables must be sized for the contingency condition (full load on one UPS) since the purpose of N+1 redundancy is to maintain full capacity during a UPS failure.

Modern server power supplies with active PFC have a cleaner harmonic profile than older passive-PFC units, but the scale of a data centre means even modest percentages produce significant currents.

Total harmonic distortion:

THD_i = √(33² + 12² + 7² + 5² + 3² + 2² + 1.5²) / 100 — (Eq. 3)

THD_i = √(1,089 + 144 + 49 + 25 + 9 + 4 + 2.25) / 100

THD_i = √(1,322.25) / 100

THD_i = 36.4%

This is moderate by data centre standards. Older facilities with passive-PFC supplies can exceed 100% THD_i.

The true RMS phase current is higher than the fundamental due to harmonic content:

I_RMS = I₁ × √(1 + THD_i²) — (Eq. 4)

I_RMS = 693 × √(1 + 0.364²)

I_RMS = 693 × √(1 + 0.1325)

I_RMS = 693 × √(1.1325)

I_RMS = 693 × 1.064

I_RMS = 737 A per phase

The harmonics add 44 A (6.4%) to the phase current beyond the fundamental. This is modest but must be accounted for in cable sizing.

In a balanced three-phase four-wire system, fundamental currents and non-triplen harmonics cancel in the neutral. Triplen harmonics (3rd, 9th, 15th, etc.) are zero-sequence and add arithmetically — multiplied by 3 — in the neutral conductor.

I_N = 3 × √(Σ I_h,triplen²) — (Eq. 5)

Triplen harmonic currents (absolute values at contingency load):

I_N = 3 × √(228.7² + 34.7² + 10.4²)

I_N = 3 × √(52,324 + 1,204 + 108)

I_N = 3 × √(53,636)

I_N = 3 × 231.6

I_N = 695 A

The neutral current (695 A) exceeds the fundamental phase current (693 A) and approaches the RMS phase current (737 A).

Neutral-to-phase ratio:

I_N / I_phase,RMS = 695 / 737 = 0.943

The neutral carries 94.3% of the phase current. If the 3rd harmonic content were 38% instead of 33% (easily possible with older PSUs or during periods of peak non-linear loading), the neutral current would exceed the phase current:

At 38% third harmonic: I_N = 3 × √((0.38 × 693)² + 34.7² + 10.4²)

I_N = 3 × √(69,327 + 1,204 + 108) = 3 × √(70,639) = 3 × 265.8

I_N = 797 A (108% of I_RMS)

The cables must be sized for the contingency RMS current of 737 A per phase. With two parallel runs per phase:

I_{per_run} = 737 / 2 = 368.5 A per run — (Eq. 6)

Ambient temperature correction (k₁):

Reference ambient = 40°C. Actual = 30°C. From Table 22, Row 30°C, Column 90°C XLPE:

k₁ = 1.08 (below reference ambient = bonus)

Grouping correction (k₂):

8 circuits on ladder tray (4 existing + 4 UPS runs, single-core trefoil groups). From Table 24, 8 trefoil groups on tray, touching:

k₂ = 0.68

Combined derating:

k_total = 1.08 × 0.68 = 0.734

Required tabulated ampacity per run:

I_z = 368.5 / 0.734 = 502 A

From AS/NZS 3008 Table 12, single-core XLPE copper on tray (trefoil):

Selected: 240 mm² per phase per run (2 runs × 3 phases = 6 phase cables, plus neutral).

Temperature correction (C_a):

Reference ambient = 30°C. Actual = 30°C.

C_a = 1.00 (no correction needed)

Grouping correction (C_g):

From Table C.3, 8 trefoil groups on tray:

C_g = 0.68

Harmonic derating (from BS 7671 Table C.4):

For THD_i between 15% and 33%, and third harmonic between 33–45% of fundamental phase current, BS 7671 Table C.4 provides a size-the-neutral-to-phase approach. The neutral must be sized as a current-carrying conductor. No additional phase derating is required if neutral is sized independently.

However, if the neutral is included as a current-carrying conductor for grouping purposes, it changes the grouping factor:

Using 4-conductor grouping:

C_g = 0.64 (8 four-core equivalent groups)

Required tabulated ampacity per run:

I_z = 368.5 / (1.00 × 0.64) = 576 A

From BS 7671 Table 4J4A, single-core XLPE on tray:

Selected: 300 mm² per phase per run.

Same methodology as BS 7671. With harmonic neutral counted:

C_g = 0.64

I_z = 368.5 / 0.64 = 576 A

From IEC 60364-5-52, Table B.52.4:

Selected: 300 mm² per phase per run.

Temperature correction: At 30°C ambient, no correction for 90°C cable.

Grouping adjustment: NEC 310.15(C)(1). With single-core cables in tray, NEC 392.80 applies. For maintained spacing (one cable diameter), no derating. If touching: 0.70.

Assume cables touching for conservative design:

C_g = 0.70

Continuous load factor: UPS output is continuous load per NEC 210.19.

F_cont = 1.25

Required ampacity per run:

I_z = (368.5 × 1.25) / 0.70 = 460.6 / 0.70 = 658 A

From NEC Table 310.16, 90°C column:

Selected: 600 kcmil (304 mm²) per phase per run.

The neutral must carry 695 A in contingency. With two parallel runs:

I_{N,per_run} = 695 / 2 = 347.5 A per run — (Eq. 7)

The neutral sizing follows the same derating methodology as the phase conductors, but the neutral is sized for its own current, not the phase current.

AS/NZS 3008:

I_z,neutral = 347.5 / 0.734 = 473 A

From Table 12: 240 mm² (538 A). Same as phase conductor.

BS 7671 / IEC 60364:

Per BS 7671 Regulation 524.2.1 and IEC 60364-5-52 Clause 524.2: where the neutral current exceeds the phase current due to harmonics, the neutral must be sized for the neutral current. Since I_N (695 A) < I_phase,RMS (737 A) in our scenario:

Neutral ≥ phase conductor = 300 mm² per run

NEC:

NEC 310.15(E) addresses neutral conductors carrying non-linear load currents. The neutral is considered a current-carrying conductor.

I_z,neutral = (347.5 × 1.25) / 0.70 = 620 A

From Table 310.16: 500 kcmil (253 mm²) (620 A). Slightly smaller than the 600 kcmil phase conductors.

For the 35 m route at contingency current:

240 mm² parallel pair (AS/NZS), I = 693 A fundamental, PF = 0.95:

From AS/NZS 3008 Table 35, 240 mm², single-core trefoil:

r = 0.091 mΩ/m, x = 0.080 mΩ/m

Per parallel pair (impedance halved): r_eff = 0.091 / 2 = 0.0455 mΩ/m

ΔV = √3 × 693 × 35 × (0.0455 × 0.95 + 0.040 × 0.312) / 1000 — (Eq. 8)

ΔV = 42,008 × (0.0432 + 0.0125) / 1000

ΔV = 2.34 V = 0.59%

Within the AS/NZS 3000 limit of 5%. PASS.

300 mm² parallel pair (BS 7671), I = 693 A, PF = 0.95:

r_eff = 0.073 / 2 = 0.0365 mΩ/m

ΔV = 1.732 × 693 × 35 × (0.0365 × 0.95 + 0.040 × 0.312) / 1000

ΔV = 1.98 V = 0.50%. PASS.

600 kcmil parallel pair (NEC), I = 693 A, PF = 0.95:

r_eff = 0.070 / 2 = 0.035 mΩ/m

ΔV = 1.732 × 693 × 35 × (0.035 × 0.95 + 0.040 × 0.312) / 1000

ΔV = 1.92 V = 0.48%. PASS.

All voltage drops are well under 1% — expected for the short 35 m route length. Voltage drop is not the governing factor here.

The neutral current is highly sensitive to the third harmonic percentage. The following table shows how neutral current changes with server fleet composition:

This table is the core engineering message: the neutral-to-phase current ratio is a direct function of the third harmonic percentage, and the crossover point where neutral exceeds phase occurs at approximately 38% third harmonic content. Any data centre with mixed-age servers or significant non-linear lighting loads will approach or exceed this threshold.

The key finding in this example is that standard cable current rating tables assume the neutral carries negligible current in a balanced three-phase system — an assumption that is completely wrong for data centre loads. The triple-N (triplen) harmonics — 3rd, 9th, 15th — do not cancel in the neutral conductor. They add, multiplied by three, because they are zero-sequence components that are in phase across all three phases.

At just 33% third harmonic content (typical for modern active-PFC server power supplies), the neutral carries 94% of the phase current. At 38%, it exceeds the phase current. At 80% (older passive-PFC supplies still found in legacy equipment), the neutral current is 2.27 times the phase current.

This creates a hidden failure mode: an engineer who sizes the neutral equal to the phase conductors (standard practice for linear loads per AS/NZS 3008 Table 10, BS 7671 Regulation 524.1, IEC 60364-5-52 Clause 524.1) will have a neutral that is technically adequate for this specific scenario (695 A < 737 A). But the margin is razor-thin — a 5% increase in third harmonic content pushes the neutral over the edge.

The engineering solution is simple: size the data centre neutral conductor for at least 173% of the phase conductor cross-section (covering up to 60% third harmonic content), or better yet, perform a harmonic analysis of the actual load profile and size accordingly. The cost of oversizing the neutral by 50–70% is trivial compared to the cost of a neutral burnout that takes the data centre offline.

BS 7671 is the only standard among the four that explicitly addresses this through Table C.4, which provides harmonic reduction factors for cables serving non-linear loads. The other standards require the engineer to calculate the neutral current from first principles — a calculation that is routinely skipped in practice, with catastrophic consequences.