Power_
Power-On Sequencing for a GPU Hall: Inrush, Soft-Start, and Staged Energization
Details a field-proven staged power-on sequence for GPU halls that limits simultaneous inrush from thousands of PSUs, prevents upstream breaker trips, and integrates with rack-level soft-start hardware during commissioning.
Key facts
- NEC Article 645 covers IT equipment power and requires documented selective coordination of overcurrent devices.
- GPU rack PSUs present inrush dominated by input capacitor charging that must be managed before full bus voltage application.
- Soft-start modules ramp PSU input current over a controlled interval set by the OEM power supply specification.
- Staged sequences energize one row or zone at a time with interlocks tied to upstream breaker status and load metering.
- IEEE 242 provides recommended practice for protection and coordination studies that include inrush and motor-start curves.
- Field crews verify each stage with calibrated clamp meters and power quality recorders before releasing the next zone.
- Leviathan Systems performs these sequences as part of final commissioning on NVL72-class deployments.
Assessing Inrush Characteristics of GPU Rack PSUs
Each GPU rack contains multiple power supplies whose input stages contain large bulk capacitors. These capacitors produce a high-magnitude, short-duration current spike when voltage is first applied. The spike magnitude and duration depend on the AC line impedance, the specific PSU model, and the temperature of the capacitors at the moment of energization.
Field crews obtain the inrush envelope from the OEM power supply data sheet and confirm it against measured values on a sample rack before full hall rollout. This envelope is then used to size the allowable number of racks that may start together without exceeding the instantaneous trip setting of the upstream breaker. Without this step, simultaneous closure of hundreds of contactors can exceed the magnetic trip threshold even when steady-state load remains well below the breaker rating.
Rack-Level Soft-Start Implementation
Soft-start contactors or solid-state modules are installed between the rack PDU and the individual PSUs. These devices limit di/dt by inserting resistance or using phase-angle control during the first several hundred milliseconds. The ramp time is set to the shortest interval that keeps peak current below the coordination study limit while still allowing the PSU to reach full output voltage before the next rack is released.
The modules are interlocked so that a fault in one rack does not prevent the remainder of the sequence from continuing. After each rack reaches steady state, the soft-start bypass contactor closes and the control system records voltage, current, and power factor before signaling readiness for the next stage.
Row and Zone Sequencing Logic
The overall hall is divided into zones sized so that the inrush of one zone plus the running load of previously energized zones stays below the available fault current and breaker settings. Sequence timers or PLC logic enforce a minimum dwell period between zones, long enough for capacitor pre-charge currents to decay and for the upstream breaker to reset its thermal memory.
Zone boundaries align with physical row or aisle layouts so that cabling and cooling loops can be verified independently. The sequence begins at the most downstream zone and works upstream toward the main switchgear, allowing any voltage drop or harmonic interaction to be observed before the next block is added.
Upstream Breaker Coordination and Settings Review
A coordination study performed to IEEE 242 guidelines sets the long-time, short-time, and instantaneous pickup points so that rack-level faults clear locally while the main breaker remains closed during staged inrush. The study must account for the reduced available fault current when only a portion of the hall is online.
Before first power-on, the settings are verified against the actual installed breakers and any adjustable trip units are locked after calibration. Changes to upstream utility feeder impedance or added generator capacity require the study to be rerun before the next major deployment.
Verification Measurements and Documentation
After each zone is released, crews record three-phase voltage, current, and total harmonic distortion at the rack PDU and at the upstream switchgear. These values are compared against the coordination study predictions; any deviation beyond the tolerance band halts the sequence until the cause is identified.
All measurement data, breaker settings, and sequence timestamps are archived with the commissioning package. This record becomes the baseline for troubleshooting later capacity additions or after utility events that may have altered source impedance.
Common Failure Modes Encountered in the Field
The most frequent failure is an upstream breaker trip during the second or third zone because the coordination study used nameplate inrush values that were lower than actual measured values at the site ambient temperature. Another common issue occurs when soft-start modules are bypassed too early, allowing residual capacitor charging current to combine with the next zone's inrush.
Miswired interlocks between zones can also cause simultaneous energization when a single rack fault should have isolated only that rack. These problems are caught by requiring a manual verification step after each zone and by comparing recorded waveforms against the study curves before proceeding. Crews that skip the verification step routinely discover the error only after the main breaker has already opened.
Standards referenced: NEC Article 645 · IEEE 242
Frequently asked_
How many racks can be started in one zone without exceeding breaker limits?
The exact count is determined by the coordination study that incorporates measured inrush data, source impedance, and the specific breaker trip curves. Crews start with a conservative single-rack test, then add racks while monitoring peak current at the upstream device. The final zone size is documented and locked into the PLC sequence.
What happens if a soft-start module fails during the sequence?
The interlock prevents the bypass contactor from closing and the zone controller skips that rack while continuing with the remaining racks. A fault alarm is raised and the rack is isolated for inspection before the next full-hall power cycle. This prevents a single module failure from cascading into an upstream trip.
Do we need to re-run the coordination study after adding more racks?
Yes. Any change in connected load or source configuration requires an updated study because the available fault current and thermal memory of the breakers will shift. Leviathan Systems includes this update as part of the change-order process for capacity expansions.
Which measurement points are required at each stage?
Voltage and current are recorded at the rack PDU input, at the row-level distribution board, and at the main switchgear. Power quality recorders capture the waveform for at least ten cycles after each contactor closure. These points allow immediate detection of unexpected voltage drop or harmonic amplification.
How long should the dwell time be between zones?
The dwell must exceed the longest expected capacitor pre-charge time plus the breaker thermal memory reset interval given in the coordination study. In practice this is verified by confirming that current has returned to steady-state values and that the upstream breaker electronic trip unit shows no pending thermal alarm before the next zone is released.