Data center rack PDU types and power distribution field guide

What a rack PDU is and where it sits in the power chain

A rack PDU is the power strip inside a server cabinet. It takes power from the floor PDU, the RPP, or the overhead busway and splits it into the outlets the servers' power supplies plug into. That makes it the last piece of the power chain before the IT load. Everything upstream, the utility, the generators, the UPS, the floor PDU, exists to get clean power to this strip, and this strip is what hands it to the gear.

Keep the two PDUs straight, because the same three letters get used for both. The floor PDU is the transformer-based cabinet out on the floor, covered in the PDU and RPP commissioning guide. The rack PDU is the strip inside the cabinet, and that is what this guide is about. When a drawing says PDU, find out which one before you order anything or quote anything.

The strip looks simple, and it is not where most people expect trouble. The trouble is choosing the wrong intelligence level and stranding capacity, sizing the input too small for the rack that fills in two years, or running A and B strips that are not actually independent. Those calls all get made before the cabinet is loaded, when changing them is cheap.

What is the difference between a rack PDU and a floor PDU?

The floor PDU is the big distribution cabinet on the floor; the rack PDU is the strip inside the rack. The floor PDU, sometimes drawn just as PDU, takes UPS output, usually steps it through a transformer to the rack voltage, and feeds panelboards or subfeeds that run out to the rows. It is floor-standing gear with a transformer, a main, and breakers. The rack and cabinet selection guide and the PDU and RPP commissioning guide cover that end.

The rack PDU does none of the transforming. It is a power strip, 0U vertical or horizontal, that takes one branch circuit from the floor PDU, the RPP, or a busway tap and breaks it into outlets for the servers. No transformer, no bulk distribution, just the last split to the gear.

The naming collision causes real mistakes. A spec that calls for PDU monitoring might mean branch monitoring at the floor PDU or per-outlet metering at the rack strip, and those are different scopes and different budgets. Read which device the requirement attaches to. The floor PDU feeds the rack PDU feeds the server, in that order, and the word PDU alone does not tell you which link you are looking at.

The PDU intelligence ladder: basic, metered, switched, intelligent

Rack PDUs sort into four intelligence levels, and the level you pick decides what you can see and control at the rack for the life of the room. The ladder runs basic, metered, switched, and intelligent, and each rung adds something the rung below cannot do. Pick too low and you are blind to the load or stuck sending a tech to the cabinet to cycle a hung server. Pick higher than the room needs and you pay for data nobody reads.

Basic is outlets and a breaker, nothing else. Metered adds a current readout, either at the inlet for the whole strip or per outlet. Switched adds remote on and off control of each outlet. Intelligent, sometimes called managed or monitored, is the networked strip that combines per-outlet metering, switching, and environmental sensing and reports all of it to the network. The line between them is what data the strip collects, whether you can reach it over the network, and whether it can switch an outlet without a person at the rack.

The table lines the four up against what each one adds and where it earns its cost. The sections after it take metered, switched, and intelligent one at a time.

The metered PDU and rack capacity

The number one reason to buy a metered strip is capacity. A metered rack PDU shows the current it is carrying, so you can see the rack load against the strip's breaker rating before you add the next server, instead of finding the limit when a breaker trips. Inlet metering shows the total for the whole strip, the cheapest way to answer the one question that matters most: how much room is left on this rack.

Without that number, capacity is a guess, and the guess goes wrong in both directions. Guess high and you leave the rack half empty, stranding power the building paid for. Guess low and you trip the breaker on a Tuesday afternoon when the last server spins up its drives. Stranded capacity is the quiet money loss in a data center, and a metered strip is the cheapest instrument that finds it.

Outlet metering goes a step further and reports each outlet, so you can see what a single server draws and bill or charge back by device in a shared hall. That is the difference between knowing the rack is at 4 kW and knowing which two servers are eating half of it. For a colocation operator, per-outlet metering is often the line item the contract is written around.

The switched PDU and remote outlet control

A switched rack PDU lets you turn each outlet on and off over the network, and the headline use is the remote reboot. A server hangs hard at 2 a.m. and no console will reach it. With a switched strip you cycle its outlet from anywhere and the box comes back, instead of dispatching a tech across town or across a campus to push a button. On a lights-out site or a remote edge cabinet, that one capability pays for the strip.

Switching also sequences and locks. You can power up a rack in order so the storage comes alive before the servers that mount it, and you can stagger the start so a whole rack does not hit inrush at the same instant. You can disable an unused outlet so nobody plugs an unsanctioned device into a billed circuit, and you can hold an outlet off until the gear on it is ready.

The caution is the same as any remote control over critical load: the outlet you can turn off from a laptop is the outlet someone can turn off by mistake. Lock down who has switching rights, label the outlets to the gear, and treat a remote power-off of a production server with the same care as throwing its breaker by hand. The convenience and the hazard are the same feature.

The intelligent or managed PDU

An intelligent rack PDU, also called a managed or monitored strip, is the networked unit that does all of it: per-outlet metering, per-outlet switching, and environmental sensing, reported over the network to a management system. It reads amps, volts, kilowatts, and kilowatt-hours, it switches and meters each outlet, and it takes plug-in sensors for temperature and humidity at the rack. The data goes up to DCIM, where it joins the floor PDU branch monitoring and the cooling picture on one dashboard.

This is the strip a modern high-density or colocation hall standardizes on, because the data is what runs the floor. Per-outlet energy supports chargeback and efficiency tracking. The amp and phase readings drive capacity and balance decisions. The temperature and humidity sensors catch a cooling problem at the rack before it cooks the gear, and the alerts say which rack, not just that something is wrong somewhere.

The trade is cost and complexity. An intelligent strip costs several times a basic one, it has a network port that has to be secured and patched like any other device on the management network, and it generates data that is only worth the price if someone uses it. Buy the intelligence the operations plan will use, not the longest feature list on the cut sheet.

What is a 0U PDU?

A 0U PDU is a rack PDU mounted vertically in the rear channel of the cabinet, where it takes up no rack-unit space from the 19 in mounting field. The 0U means zero U: it lives in the side channel beside the gear instead of eating a slot a server could use. In a server cabinet that is the default, because rack units are the scarce thing and a vertical strip can carry 30, 40, or more outlets up the full height of the rack.

The alternative is a horizontal PDU, a 1U or 2U strip that bolts into the mounting field like any other gear and carries a smaller outlet count, commonly 6 to 16. Horizontal strips fit two-post and network racks that have no rear channel to hold a vertical, and they suit low-outlet or low-power racks where giving up a U or two does not hurt. The rack and cabinet selection guide covers which cabinets have the 0U channel and the depth a vertical strip needs.

The other reason 0U vertical wins in a server cabinet is heat. The rear channel runs hot off the gear exhaust, and a 0U strip is built for it, with components rated for the higher temperature back there. A horizontal strip sitting in the airflow path is one more thing to cool. Confirm the cabinet has the channel and the depth for the vertical before you spec it, because a 0U strip with nowhere to mount ends up stealing the rack units it was supposed to save.

Single-phase versus three-phase rack PDUs

A single-phase rack PDU carries one phase to the rack; a three-phase strip carries all three, and the three-phase strip puts far more power in the same cabinet. The rule of thumb the industry uses is that racks below roughly 5 kW do fine on single-phase, and above that three-phase is how you land the kilowatts in the cabinet without an absurd number of strips and circuits. A three-phase strip carries more power per PDU, which is the whole point on a dense rack.

The voltage moves with it. North American three-phase rack power is commonly 208V, and the higher-density move is to 415V three-phase, often written 400/415V, which delivers 240V phase-to-neutral at the outlet. Going to 415V roughly doubles the cabinet's power capacity over 208V at the same amperage, without doubling the conductor and connector count, which is why high-density and AI rooms lean on it. The IT gear still sees 208V or 240V at the cord; the higher voltage rides on the strip's input.

The catch with three-phase is that you now have to keep the three phases balanced, which the next section covers. A single-phase strip cannot be out of balance because there is only one phase. A three-phase strip hands you more power and the job of spreading the load across it evenly.

Phase balancing on a three-phase strip

Phase balancing on a three-phase rack PDU is spreading the single-phase loads evenly across the three phases so no one phase hits its limit while the other two coast. The strip's total rating assumes the load is shared. Pile the servers onto one phase and that phase trips or strands capacity while the cabinet still reads half empty on the others. The power is there. The imbalance makes part of it unreachable.

The old way to balance was to assign outlets by branch and count amps as you loaded the rack, keeping each phase within about 10 percent of the others. That works, but it drifts, because the load grows one server at a time after turnover and nobody recounts. Many modern three-phase strips alternate the phase outlet by outlet, so consecutive outlets land on different phases and the rack stays close to balanced as it fills, without a person tracking it. Confirm whether the strip alternates phases or groups them by branch, because it changes how you assign the gear.

Imbalance is not just a capacity problem. In a wye system the neutral carries the difference between the phases, so a badly unbalanced strip pushes current onto the neutral on top of the harmonic current the IT load already puts there. Balance the phases and the neutral settles down. Let it drift and you stack imbalance on harmonics on a neutral that is working hard already. The 10 percent figure is a common operational target; the strip's ratings and the design control the real limit.

The input: plug, cord, and voltage

The input is the plug, cord, and voltage that feed the strip, and it has to match the receptacle on the whip from the floor PDU, the RPP, or the busway tap. For higher-amp three-phase strips the input is commonly an IEC 60309 pin-and-sleeve connector, the locking plug sized for the amperage and phases, in the 30A and 60A range that high-density racks pull. Lower-power single-phase strips often take a NEMA locking plug, the twist-lock L-series, or an IEC inlet. The connector is what physically lands the strip on its circuit, and it has to agree with what the electrician put on the whip.

Input amperage sets the strip's ceiling, and it is rated before any continuous-load derating. A 30A input is not 30A of usable continuous load. The capacity and breaker section covers how the 80 percent rule eats into that number. Size the input to the rack the cabinet will hold in a couple of years, not the half-empty rack on day one, because changing the input means a new strip and a new whip pulled to the cabinet.

Where the whip comes from is a design call. A fixed floor PDU or RPP runs a whip per rack, sized and breakered for the strip it feeds. An overhead busway lets you add or move a tap box as the row fills, which is why high-density and build-as-you-grow rooms use it, and the PDU and RPP commissioning guide covers the floor end of the whip and the busway tap. Confirm the receptacle type, the voltage, the phases, and the amperage of the whip before the strip shows up. A strip whose plug does not match the whip is a strip that does not get installed that day.

The outlets: C13, C19, count, and locking

The outlets are where the servers plug in, and on a data center rack PDU they are almost all IEC 60320: C13 for the common server and appliance cords, and C19 for the heavier gear that pulls more current, such as blade chassis and large switches. A strip's outlet count and the mix of C13 to C19 have to match the cords in the cabinet. A rack full of dual-power-supply servers needs enough outlets on each of two strips to land every cord, and a rack with a few big chassis needs the C19s to carry them.

Count the outlets against the gear and the redundancy, not the slot count of the rack. A 42U rack of 1U dual-corded servers can need 40-plus outlets per side, which is exactly why 0U vertical strips run so long. Run out of outlets and the fix is a second strip per side or a re-cable, neither of which is free once the rack is in production.

The outlets worth paying for are locking. A standard C13 cord can back out of its outlet from vibration, a cable bump during a move next door, or its own weight on a vertical strip, and a server that loses its cord goes down for no reason anyone can see at the screen. Locking outlets, the kind that grip the cord until you release a latch, stop that. On A and B strips feeding dual-corded gear the cost is cheap insurance, because a bumped cord on a single-corded device is an outage. Many strips offer locking outlets as a build option; confirm it on the order.

The A and B feeds and dual-corded servers

The A and B feed is how a rack gets redundant power: two rack PDUs, an A strip and a B strip, fed from two independent paths, so a dual-corded server takes one cord from each. Lose one side and the server rides on the other. Almost all enterprise and colocation server gear ships with two power supplies for exactly this, and the two strips are what those two supplies plug into.

The point of A and B is that the two paths are genuinely separate, all the way back. The A strip and the B strip have to trace to different floor PDUs, different UPS systems, ideally different generator and utility paths. The PDU and RPP commissioning guide covers tracing the A and B feeds back to the source. At the rack, the A and B discipline is physical: A strip one color, B strip another, A cords to the A side, B cords to the B side, every cabinet the same way, so a tech can see the split without a meter.

The failure to hunt for is the A and B that are not. Two strips that look independent at the rack but trace back to the same floor PDU or the same upstream breaker are one feed wearing two cords, and they drop the rack together the day that source fails. The label says A and B. The whips and the one-line have to agree that they really are two paths, or the redundancy is a paint job.

The redundancy budget and the static transfer switch

The trap in A and B power is loading. If both strips run near full and one side fails, the whole rack lands on the survivor, and the survivor trips. The redundancy you paid for then takes the rack down instead of saving it. To ride through a side failure, each strip has to carry no more than what one strip can hold alone, which in a two-side setup means loading each side to roughly half its capacity. That looks like wasted headroom right up until the day it is the only thing keeping the rack up.

This is the budgeting people get wrong. A metered strip on each side is what lets you watch the per-side load and hold the failover margin, which is a strong argument for metered or better on any A and B rack. Run them blind and the rack creeps past the failover limit one server at a time, and nobody knows until a side drops and takes everything with it.

Single-corded gear does not fit the A and B model on its own, because it has one cord and one strip. The fix is a rack-mount static transfer switch that takes both an A and a B feed and hands one output to the single-corded device, switching to the surviving source if one fails. The static transfer switch is what brings a single-corded box into the redundant rack. Where the gear is dual-corded, the two power supplies do the job and no transfer switch is needed.

Capacity, branch breakers, and the 80 percent rule

The strip's capacity is set by its input rating and its branch breakers, and the number you can actually use is less than the nameplate because of continuous-load derating. The NEC and UL convention treats a load running three hours or more as continuous and limits it to 80 percent of the breaker rating. So a 30A strip holds about 24A of continuous IT load, not 30A, and a 20A branch holds 16A. Treat the nameplate as the ceiling and the 80 percent figure as the working limit, and confirm it against the manufacturer's listing and the adopted code, because the strip is built and listed around that derating.

Bigger strips break the input into branches, each with its own breaker, so a fault or overload on one branch trips that branch and leaves the rest of the strip up. UL has required branch-circuit protection inside the PDU where the inlet current exceeds the outlet current, which is why larger strips carry their own breakers. Spreading the gear across branches keeps one heavy device from monopolizing a branch and keeps a single trip from dropping the whole strip. Know how many branches your strip has and which outlets belong to which branch, because an outlet group that all hangs on one branch breaker fails together.

The blunt version: do not load a strip or a phase to its nameplate. Hold continuous load to 80 percent of the breaker, balance it across the phases and the branches, and on A and B racks hold each side to its failover budget on top of that. The strip that runs at its rated number is the strip that trips on the hot afternoon when the load comes in heavier than the schedule said.

High-density and AI rack power

High-density racks, the ones built for AI and GPU compute, have rewritten what a rack PDU has to carry. A classic server rack was planned around a few kW. An AI rack can pull well past 50 kW, and the largest builds are reaching for far more, which forces three-phase strips, high input amperage of 60A and 100A and up, and the higher 415V distribution to fit the power in the cabinet at all. On these racks three-phase is not a choice. It is the only way to land the kilowatts.

The feed usually comes from busway rather than a fixed whip, because a dense, fast-changing room needs to add and move taps as the racks land, and the strips themselves are intelligent because a rack drawing this much power cannot be run blind. Per-outlet metering, phase data, and environmental sensing are how a 50 kW-plus rack is operated safely. The power and the data scale together.

This is the fastest-moving part of the topic, and the numbers climb every year, so treat any specific kW or amperage figure as a snapshot and size to the actual gear and the project's growth plan. The constant is the direction: more power per rack, three-phase, higher voltage and amperage, busway feeds, and intelligent strips. The rack and cabinet selection guide covers the cabinet, weight, and liquid-cooling side of the same high-density rack.

How does rack PDU monitoring and alerting work?

Rack PDU monitoring is the intelligent strip reporting its readings over the network and raising an alert when a reading crosses a threshold you set. The strip measures current, voltage, power, and energy, per strip and often per outlet, and it watches plug-in sensors for temperature and humidity at the cabinet. Those readings feed DCIM or a building management system, where they sit alongside the floor PDU branch monitoring and the cooling data.

The alerts are what make the monitoring worth anything. Set a warning and a critical threshold per strip and per phase against the breaker rating, so operations gets told before a strip reaches its limit, not after it trips. Set a phase-imbalance alert so a drifting rack flags itself. Set temperature and humidity thresholds at the rack so a cooling failure shows up as an alert naming the cabinet, not as dead gear an hour later. Monitoring with no thresholds is a log nobody reads.

The discipline that makes it useful is the same one that makes branch monitoring useful: the data has to map to the steel. The strip in DCIM has to be tied to the right rack, and the outlet readings to the right gear, or operations gets confident, precise, wrong numbers. Map it, set the thresholds to the ratings, and the strip becomes an instrument instead of a feature on the cut sheet.

Installing and cabling the rack PDU

Installing a 0U strip is mostly mounting and matching. The vertical strip snaps into the accessory rails in the rear channel, usually tool-less, on the side the A or B feed lands. Confirm the cabinet has the channel and the depth for the strip with the gear and cable already in, because a strip that fits an empty cabinet can foul the rear door once the servers and their cords go in behind it. Land the input whip, confirm the plug matches the receptacle, and verify the voltage and phases before you energize.

The cabling is where a rack stays sane or turns into a trap. Route the cords so they do not cross the airflow path or block the rear door, keep the A cords on the A strip and the B cords on the B strip, and color-code the two sides so the split is visible at a glance. Use the cord locks if the strip and cords support them, because a dropped cord is an outage with no error message. The rack and cabinet selection guide covers the cable management hardware and the channel the cords route in.

Then assign and label. Map each outlet to the gear on it, label the outlet, the cord, and the gear to one naming scheme keyed to the rack coordinate, and write the assignment into the record and the DCIM. A strip whose outlets nobody labeled is a strip where every move turns into a trace-it-out on a live rack, which is where someone unplugs the wrong server. Label as you install, not after the rack is full, because a label filled in from memory later is a guess.

How do I choose a rack PDU?

Start with what the room needs to see and control, then size the power. The intelligence level is the first call: basic where metering lives upstream and the rack is low-change, metered where you have to track capacity, switched where you need remote reboot or outlet control, intelligent where DCIM runs the floor. Most data center server racks land on metered at a minimum, and high-density or colocation racks land on intelligent.

Then the power: single-phase below roughly 5 kW, three-phase above it, at the voltage and input amperage the rack density needs, with the input plug matched to the whip. Then the form factor, 0U vertical for a server cabinet with a rear channel, horizontal for a two-post or low-outlet rack. Then the outlets, enough C13 and C19 in the right mix for the gear and the redundancy, with locking outlets where a dropped cord is an outage. And A and B strips for any rack that has to ride through a feed failure.

Two strips with the same outlet count can be wrong for opposite reasons, one blind where you needed metering and one single-phase where the density needed three. Match the strip to the rack and the room together, and let the project specification and the manufacturer's ratings control the final numbers.

Rack PDUs at data-center scale

One rack PDU is simple. A data center has hundreds or thousands of them, and at that count the problem changes from picking a strip to standardizing one. A hall that buys one or two strip models across the floor can stock spares, train techs once, and roll the monitoring data up the same way for every rack. A hall that bought whatever was cheapest rack by rack has a parts bin of one-offs and a monitoring map full of exceptions.

The standardization is what makes the data usable. When every rack reports the same readings in the same format, DCIM can roll the whole floor into one capacity and power picture, and operations can compare rack to rack and find the stranded capacity and the drifting phases. A floor of mismatched strips reports a mess nobody trusts, which is the same as no data.

Set the standard at design, buy to it, and hold the line as the floor grows. The strip is a small line item next to the gear it feeds, but multiplied across a hall it is the instrument the whole floor's power is run from, and the consistency is worth more than the unit price.

What to document

The rack PDU record is what operations runs the rack against and what the next person reads before they add a server. Capture it per strip, keyed to the rack coordinate, so the strip, its ratings, its feed, and its outlet assignments are on file before the gear lands.

Record the strip type and intelligence level, the input plug, voltage, phases, and amperage, the source whip or busway tap it lands on and whether it is the A or B side, the outlet count and mix, the per-phase and per-side loading, the outlet-to-gear assignments, and the monitoring map and thresholds. The point is that a reviewer can confirm the rack's power from the record alone, without tracing cords on a live rack.

Common mistakes

• Specifying a basic strip where the rack needed metering, so capacity is a guess and the breaker trips or sits stranded.
• Overloading a strip or a phase past the 80 percent continuous limit on the breaker rating.
• Loading both A and B strips near full, so one side failing drops the rack onto a survivor that trips.
• Leaving the three phases out of balance, so one phase trips while the other two coast.
• Running A and B strips that trace back to the same floor PDU or upstream breaker.
• Skipping locking outlets, so a bumped or sagging cord drops a server with no error at the screen.
• Under-sizing the input for the rack the cabinet will hold in two years, forcing a strip and whip swap later.
• Mounting a horizontal strip in the airflow path where a 0U vertical belonged, stealing rack units and blocking air.
• Energizing intelligent strips without mapping the monitoring to the rack or setting alarm thresholds.
• Labeling the outlets from memory after the rack is loaded, so the assignments and the gear disagree.

Field checklist

Standards and references

The rack PDU's electrical ratings come from the manufacturer and its UL listing, and the strip is built and listed around the NEC and UL continuous-load convention that limits a continuous load to 80 percent of the breaker rating. Treat the input amperage and the branch ratings as the manufacturer's numbers for that model, and the 80 percent derating as the working limit, confirmed against the listing and the adopted NEC, NFPA 70, edition. UL has required branch-circuit protection inside the PDU where the inlet current exceeds the outlet current, which is why larger strips carry their own branch breakers.

The outlet and inlet connectors follow IEC 60320, the C13, C19, C14, and C20 family for the cords and inlets, and the higher-amp three-phase inputs commonly use IEC 60309 pin-and-sleeve connectors. NEMA locking configurations cover many single-phase inputs in North America. These are connector standards; confirm the exact type against the strip and the whip.

For the room around the strip, TIA-942 frames the data center infrastructure, ASHRAE TC 9.9 gives the thermal guidelines the rack environment is held to, and the Uptime Institute Tier framework drives the A and B redundancy where a Tier is claimed. DCIM is the management layer the intelligent strips report into. Edition letters, amperage ratings, and connector types move between products and code cycles, so confirm every number against the manufacturer's data and the adopted standard before citing it on a submittal.

Units, terms, and acronyms

The rack PDU carries its own vocabulary, and the same strip gets described several ways across a cut sheet, a one-line, and a DCIM screen. Power per rack is stated in kilowatts, current in amps, and energy in kilowatt-hours. Rack height is in rack units, where a 0U strip takes none. Input and outlet types are named by their connector standard. The terms below travel across the selection, the install, and the record.

Rack PDU

The power strip inside a cabinet that distributes a branch circuit to the gear's power supplies

0U / vertical

A strip mounted in the rear channel that takes no rack-unit space from the 19 in field

Metered / outlet-metered

A strip that reads total current at the inlet, or current per outlet

Switched

A strip whose outlets can be turned on and off remotely over the network

Intelligent / managed

A networked strip with metering, switching, and sensors reported to DCIM

A and B feed

Two strips on two independent paths for redundant power to one rack

C13 / C19

IEC 60320 outlets for standard and higher-current server cords

IEC 60309

The pin-and-sleeve locking input connector for higher-amp three-phase strips

Phase balance

Spreading load across the three phases so no single phase overloads

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What is the difference between a PDU and an RPP in a data center?
• What is a rack unit (U)?
• How does power get from the utility to a server?
• Why are data centers moving to DC power for AI racks?
• What kind of batteries do data centers use?