Rack Power Delivery and Busways
Rack power determines current, distribution equipment and redundancy requirements. Training workloads can also create load swings that average power hides.
TL;DR: For a three-phase feed, current depends on power, voltage and power factor. A 132 kW load at 415 V and a power factor of 0.99 draws about 185 A. Each redundant feed must support the required failover load. Measure transients and follow the system’s electrical requirements; a software GPU power cap is not a substitute for electrical protection.
Two terms first. Three-phase power delivers electricity on three conductors whose cycles are offset, which carries more power per conductor than a single phase; that is why large racks use it. The power factor is the fraction of the supplied current doing useful work, so 0.99 is close to ideal. A PDU (power distribution unit) is the strip inside the rack that splits the feed to each server.
The arithmetic that sets the design
current from power, three-phase
I = P / (sqrt(3) x V x power factor)
at 132 kW, 415 V, power factor 0.99:
I = 132,000 / (1.732 x 415 x 0.99) = 185 A
what that means for distribution
a common rack PDU tops out well below this, so one strip cannot feed the rack
the answer is either several PDUs per rack or a busway tap rated for the load
and for redundancy the facility provisions A and B feeds each able to carry the whole rack,
so each path must support about 185 A before required design margins;
the normal total load remains 132 kW, and billing depends on the contract
the same rack on other voltages
at 240 V single phase: 132,000 / 240 = 550 A, which is not practical
at 480 V three-phase: 132,000 / (1.732 x 480 x 0.99) = 160 A
sanity: higher voltage is how the current stays manageable, which is why dense halls run
415 V or 480 V distribution rather than the 208 V that older halls were built for
What a rack actually draws, by generation
| Configuration | GPUs per rack | Approximate rack power |
|---|---|---|
| Example older rack | configuration-dependent | calculate from the installed servers |
| 4x DGX B300, per NVIDIA's reference architecture | 32 | over 50 kW |
| GB300 NVL72 | 72 | about 120 kW by NVIDIA's figure, 132 to 142 kW nominal with peaks near 155 kW per integrator deployment guides |
The range on the last row is worth carrying around exactly as a range with both attributions, because the two numbers answer different questions. NVIDIA's figure describes the system; the integrator figures describe what a commissioned rack pulls in a real hall with its own networking, storage adapters and margins. Power and Datacenter Constraints covers the facility side of this; the rack-level point is that the difference between 120 and 155 kW is larger than an entire rack of the previous generation.
The transient, and why it is a real problem
Training phases can make many devices change load together. Communication may overlap compute, so neither the size nor timing of the swing is universal. Measure the rack and upstream response using representative jobs.
A training phase changes device demand. The electrical question is how that change travels through the complete rack and upstream distribution.
- Representative job
Include the compute and communication phases that change demand.
- Rack measurement
Record load over time, including networking and other rack equipment.
- Distribution response
Compare measured peaks and rate of change with the supported electrical design.
- Supported control change
Test a cap or smoothing policy for both electrical response and useful throughput.
Size the distribution and protection to the supported system requirements. GPU power caps may reduce consumption, but their effect on transients and throughput needs testing. A board-level cap does not control all rack loads, and no fixed multiplier such as 1.4 applies to every installation.
Redundancy and how it is actually consumed
- A and B feeds from separate upstream paths, each sized for the full rack, so losing one path does not drop the rack. The contract determines whether and how that redundant capacity is billed.
- N+1 at the PDU and busway tap level, so a single tap failure does not take the rack.
- Dual power supplies per node, one on each feed, which is where the redundancy actually terminates.
- The failure mode to plan for includes a feed failing or being taken out for maintenance while the other is at full load, which is the moment a rack discovers it was oversubscribed.
What interviewers are listening for
Show the current calculation with voltage and power-factor assumptions. Explain how the rack remains within limits during a feed failure or maintenance, and how measured load transients affect the upstream design. Keep physical redundancy separate from billing: the contract determines how reserved capacity and energy are charged.
Key takeaways
- At 415 V three-phase and a power factor of 0.99, a 132 kW load draws about 185 A.
- Each redundant path must carry its specified failover load; that does not imply twice the energy consumption or twice the bill.
- Include networking, cooling equipment and other rack loads in the power boundary.
- Training phases can create correlated load changes. Size equipment from supported system requirements and measured peaks.
- Test power caps for performance and transient behavior; retain the required electrical protection.