Understanding Rack Power Distribution Units
A rack power distribution unit (PDU) is the last active element in a data center's power chain, delivering conditioned utility or UPS power to individual servers, switches, and storage devices mounted within a rack. Choosing the right PDU tier and designing a sound feed architecture are foundational decisions that affect uptime, operational efficiency, and compliance with applicable electrical and infrastructure standards. This guide walks through the three core PDU categories and explains why dual A+B feed design is a non-negotiable requirement in serious infrastructure deployments.
The Three Core PDU Categories
Basic PDUs
A basic rack PDU is, in essence, a managed power strip engineered for rack form factors. It distributes power from a single inlet to multiple C13, C19, or NEMA outlets but provides no telemetry or remote switching capability. Basic PDUs are appropriate in cost-constrained or low-density edge locations where power monitoring is handled upstream—for example, at the branch-circuit breaker panel—and where per-outlet granularity is not required. Their simplicity reduces failure points, but the absence of metering means any energy anomaly, overload creep, or circuit imbalance goes undetected at the rack level until a breaker trips or equipment fails.
Metered PDUs
Metered PDUs add real-time current, voltage, and power measurement—typically at the inlet level and, on higher-grade models, at the individual outlet level. Inlet-metered units display aggregate load via an LCD or report it over an IP network interface, allowing facility teams to track rack-level power consumption, enforce capacity limits, and feed data into data-center infrastructure management (DCIM) platforms. Outlet-metered PDUs extend this visibility to each individual device, enabling per-server power accounting critical for chargeback models, energy benchmarking against PUE targets, and compliance with efficiency reporting obligations. In high-density GPU compute racks operating at 60 kW or more per rack—as seen in modern AI data center designs—per-outlet metering is effectively mandatory because load imbalance between phases can be severe and difficult to detect any other way.
Switched PDUs
Switched PDUs incorporate per-outlet or per-group remote on/off switching in addition to full metering. An operator can power-cycle a hung server, sequence startup loads to avoid inrush current spikes, or shed non-critical loads during a power event—all without physical access to the rack. Some models also support outlet-level current limiting and user-access controls, which map to the electrical safety principles outlined in NFPA 70E. In colocation environments or remote edge sites where hands-on access is expensive or infrequent, the ROI on switched PDUs is straightforward. The management plane of a switched PDU should always be connected to an out-of-band management network so that a production network failure does not prevent remote power control.
Comparing PDU Types at a Glance
| Feature | Basic | Metered (Inlet) | Metered (Outlet) | Switched |
|---|---|---|---|---|
| Power distribution | ✓ | ✓ | ✓ | ✓ |
| Aggregate power monitoring | ✗ | ✓ | ✓ | ✓ |
| Per-outlet power monitoring | ✗ | ✗ | ✓ | ✓ |
| Remote outlet switching | ✗ | ✗ | ✗ | ✓ |
| DCIM / SNMP integration | ✗ | Often | ✓ | ✓ |
A+B Feed Architecture: Why Dual Feeds Are Essential
Even the most capable switched PDU cannot protect against a single upstream failure if only one power feed enters the rack. Dual A+B feed design addresses this by supplying every rack from two entirely independent power paths—typically originating from separate UPS modules, separate distribution panels, and separate branch circuits. Each PDU (the "A" unit and the "B" unit) is fed from its own path, and equipment with dual power supplies connects one PSU to each PDU. If Feed A fails—whether due to a UPS fault, a breaker trip, or upstream utility loss—Feed B sustains the load without interruption and without requiring an automatic transfer switch at the rack level.
ANSI/TIA-942 formalizes this principle within its data-center infrastructure rating framework, with increasing redundancy requirements at higher rating levels. Uptime Institute's Tier III (concurrently maintainable) specification requires, among other criteria, that every load be served by multiple active distribution paths so that any single component can be removed for maintenance without causing downtime. A+B feed design is the rack-level implementation of that principle. NEC/NFPA 70 governs the physical installation—wiring methods, grounding, and overcurrent protection—that makes these dual feeds safe and code-compliant.
Designing A+B Feeds Correctly
- Separate upstream sources: A and B feeds must originate from independent UPS modules or independent UPS systems, not from two outputs of the same unit. In a design using two online double-conversion UPS units arranged N+1, each UPS module supplies one feed path.
- Independent branch circuits: Each PDU must have its own dedicated branch circuit with appropriately rated overcurrent protection per NEC/NFPA 70. Sharing a circuit between A and B PDUs defeats the redundancy entirely.
- Load balancing: Populate servers symmetrically between A and B PDUs so that if one feed fails, the surviving feed does not become overloaded. Metered PDUs are essential here—without per-PDU load visibility you cannot verify balance.
- Grounding and bonding: Both PDUs in a rack must be properly bonded per ANSI/TIA-607 and NEC/NFPA 70 to prevent ground potential differences that could damage equipment or create safety hazards.
- Surge protection upstream: Type 1 and Type 2 surge-protective devices (SPDs) installed upstream per NEC/NFPA 70 protect the PDUs and connected equipment from transient overvoltages; this is especially relevant in facilities with on-site generation, BESS, or solar integration where switching transients are more frequent.
- Single-corded equipment: Devices with only one power supply require an automatic transfer switch (ATS) or a static transfer switch (STS) at the rack or row level to benefit from A+B redundancy. Relying on a manual process to switch feeds during an outage is operationally unreliable.
A+B Feeds in High-Density AI Compute Racks
Modern GPU compute racks can draw 60 kW or more per rack, typically distributed across 3-phase circuits. In such environments, intelligent rack PDUs operating at higher voltages and amperages—consistent with the 480 V, 3-phase, 60 A per-phase designs used in advanced AI data centers—are deployed in A+B pairs with per-outlet metering enabled. This combination provides the current headroom for extreme density, the granular telemetry needed to detect phase imbalance across hundreds of GPU nodes, and the feed redundancy required to protect capital-intensive compute assets. The metering data feeds directly into DCIM tools that track rack-level PUE contribution and flag anomalies before they become outage events.
PDU Selection Checklist for Heather Technologies Customers
- Define power density per rack before selecting PDU amperage and outlet types—undersized PDUs are a common and avoidable failure mode.
- Match PDU tier (basic/metered/switched) to the operational model: remote sites and high-density compute almost always justify metered or switched.
- Specify A+B feed architecture from the outset; retrofitting dual feeds into a live data center is costly and disruptive.
- Verify that all upstream electrical work complies with NEC/NFPA 70 and that grounding meets ANSI/TIA-607.
- Confirm DCIM or SNMP compatibility so PDU telemetry integrates with existing monitoring platforms.
- For Tier III or higher facilities, validate that the A and B paths are concurrently maintainable end-to-end, consistent with Uptime Institute Tier III requirements and ANSI/TIA-942 guidance.
Selecting the right PDU tier and committing to a properly engineered A+B feed architecture are among the highest-leverage decisions in data-center design. They are also among the easiest to get wrong when treated as an afterthought. Contact Heather Technologies for product selection support and infrastructure review services tailored to your facility's power density, redundancy tier, and compliance requirements.