How to Choose the Right PDU for Your Data Centre
How to Choose the Right PDU for Your Data Centre
Power Distribution Units (PDUs) are critical infrastructure components that directly impact data centre reliability, efficiency, and operational costs. Yet many Singapore data centre operators make PDU selection decisions without a clear framework, leading to compatibility issues, power bottlenecks, and costly upgrades down the line.
The right PDU delivers precise power distribution, real-time monitoring, and scalability as your infrastructure grows. Modern PDUs offer intelligent outlet management, environmental sensors, and remote management capabilities that reduce downtime, optimize energy consumption, and provide the visibility needed to manage complex multi-rack environments effectively.
Enova Technologies | Raritan | Data Centre Power Distribution
How to Choose the Right PDU for Your Data Centre
A buyer’s framework for Singapore data centre operators | Enova Technologies Pte Ltd
PDU selection is one of the most consequential infrastructure decisions a data centre makes — and one of the most under-specified. The wrong tier limits your visibility. The wrong inlet type means a retrofit before the rack goes live. The wrong outlet count means a second PDU mid-project. This guide works through the decision in the order it matters: tier first, specifications second, deployment context third.
A Power Distribution Unit (PDU) distributes mains power to rack equipment. In the simplest form, that means a power strip on a rail. In the most capable form, it means per-outlet metering at ±0.5% accuracy, inrush current capture, residual current monitoring, remote outlet switching, environmental sensor integration, and DCIM connectivity — all in a 1U device.
The gap between those two endpoints is wide, and the right choice depends on what you are trying to do with the data and the infrastructure after it is installed. A colocation operator with 200 racks needs different things from a PDU than a single-tenant enterprise data room with 10. An AI GPU deployment at 25 kW per rack has different requirements from a legacy server environment at 4 kW per rack.
This guide explains the tier framework, the specifications that matter most, the deployment contexts that drive the decision, and the Singapore-specific factors that raise the baseline above what most international buyer guides assume.
The four tiers of rack PDU
The industry broadly organises rack PDUs into four capability tiers. Knowing which tier you need is the first question — it determines both the product range and the price band.
| Tier | Also called | What it does | Right for |
|---|---|---|---|
| 1. Basic | Passive PDU, non-metered | Distributes power. No measurement, no monitoring, no network connectivity. | Edge closets, lab benches, low-criticality environments where power draw is known and stable. |
| 2. Metered | Input-metered PDU | Measures total power at the PDU input (aggregate amps, watts, kWh). No per-outlet visibility. Network-accessible. | Small to mid-size deployments needing basic energy logging. A starting point for PUE measurement, but insufficient for rack-level capacity planning. |
| 3. Outlet-metered | Outlet-metered PDU, per-outlet metering | Measures power at each individual outlet. Can include outlet switching for remote power-cycle control. | Colocation environments with per-tenant billing, any deployment where rack-level capacity planning requires device-by-device data, most Singapore Green Mark DC applications. |
| 4. Intelligent | Intelligent PDU, smart PDU | All of Tier 3, plus: total harmonic distortion (THD), power factor, inrush current capture, residual current monitoring (RCM), environmental sensor ports, DCIM integration, and event alerting. | AI/GPU deployments, high-density environments, regulated facilities (financial institutions, government, healthcare), and any deployment requiring energy compliance reporting or advanced fault detection. |
For most data centre deployments in Singapore, Tier 3 or Tier 4 is the correct starting point. The reasons are covered in detail below, but the short version is: Singapore’s regulatory environment, colocation standards, and AI rack density all require measurement depth that Tiers 1 and 2 cannot provide.
Five specifications that determine whether a PDU fits the job
Once you know the tier, the following five specifications determine whether a specific model fits your deployment. Getting these wrong at specification stage means a return or a retrofit.
| Specification | What to check | Common mistakes |
|---|---|---|
| Input voltage and phase | Single-phase (200–240V, 1-phase) or three-phase (380–415V, 3-phase). Match exactly to your facility’s power feed. | Specifying a single-phase PDU for a three-phase feed, or vice versa. Confirm feed type with your facilities team before ordering. |
| Input current rating (kVA / kW) | The PDU must handle the peak draw of all connected devices, including inrush. For AI GPU racks: multiply nameplate draw by 1.5× as a safety margin. | Sizing to nameplate only. GPU servers inrush at 3–5× nameplate for up to 200ms at boot. A 16A single-phase PDU will trip under inrush from a single high-density GPU server. |
| Inlet connector type | IEC C20 (16A single-phase), IEC C19 (16A locking), L6-20P or L6-30P (US NEMA, common in US-spec colo), or IEC 309 / CEE (three-phase). Singapore colo facilities typically supply IEC 309 32A or 63A three-phase for high-density. | Ordering US-spec inlet (NEMA L6-30P) for a Singapore colo with IEC 309 feeds. Always confirm the rack feed type with the colo operator before purchasing PDUs. |
| Outlet type and count | C13 (10A, standard 1U/2U servers), C19 (16A, high-power servers and storage), or mixed. Outlet count should be device count + 20% buffer. Confirm C13 vs C19 mix against your server PSU specs. | Ordering all C13 outlets for a rack with GPU servers that require C19. NVIDIA H100/H200 class systems require C19 outlets at 10–16A per PSU. |
| Metering accuracy | Tier 2 PDUs typically meter at ±2–5%. Tier 4 intelligent PDUs (Raritan PX4) meter at ±0.5%. The accuracy gap matters for colocation billing and energy compliance reporting. | Specifying input-metered PDUs (±2–5% accuracy) for environments where the power data feeds Green Mark or colo billing reports. Inaccurate metering undermines both. |
Matching PDU tier to deployment type
The correct tier is driven by what happens to the power data after it is collected. A PDU that measures power but cannot route that data to anything useful is a missed specification.
Standard enterprise data room (up to 50 racks, 4–8 kW per rack). Outlet-metered PDUs with remote switching. The switching capability enables remote server restart without a site visit, which becomes operationally significant at 20+ racks. Network-accessible metering integrates with basic monitoring tools even without a full DCIM deployment.
Colocation rack (single-tenant or multi-tenant). Outlet-metered switched PDUs are the standard specification for most Singapore colocation providers. Per-outlet metering is the basis for tenant billing where power is charged separately. Colo operators at Equinix SG, ST Telemedia, and Keppel Data Centres all specify outlet-level metering for customer racks. Confirm your specific colo’s PDU specifications before purchasing — some colo operators supply PDUs; others require tenants to specify their own to an approved list.
AI and GPU-dense deployment (12–30+ kW per rack). Intelligent PDU (Tier 4) is the correct specification. The reasons are specific: GPU servers draw inrush current at 3–5× rated load at every boot event. Standard metered PDUs log this as a brief spike and return to average — the event is invisible. Intelligent PDUs with inrush capture record the actual peak at each outlet, enabling engineers to understand real peak load versus nameplate and set appropriate circuit breaker ratings. For racks at 25 kW or above, three-phase intelligent PDUs are the correct input specification.
Government, financial institution, or regulated environment. Intelligent PDUs with RCM (Residual Current Monitoring) are increasingly specified in MAS-regulated financial institution data centres and government critical information infrastructure (CII) environments. RCM detects insulation degradation before it becomes a fault, providing early warning of fire hazard risk. MAS TRM requirements for incident detection and reporting make pre-fault alerting capabilities directly relevant.
PDU tier selection: quick reference
PDU tier selection by deployment context. Singapore-specific deployments (colocation, AI, regulated) all fall into Tier 3 or Tier 4.
Why Singapore data centres need a higher baseline than most markets
Most international PDU buyer guides are written for US or European deployments and set the baseline at metered PDUs. Singapore’s regulatory, commercial, and technical environment raises that baseline for three reasons.
Green Mark DC and DC-CFA2 energy reporting. Singapore’s Building and Construction Authority (BCA) Green Mark for Data Centres requires demonstrable PUE measurement and energy monitoring at the facility level. DC-CFA2, the framework for new data centre capacity allocation, mandates a PUE of 1.25 or better and a 50% green energy target. At the rack level, producing credible PUE data requires per-rack power measurement — which means outlet-metered PDUs, not input-metered. The 2024 Green Mark refresh explicitly favours operators who can demonstrate efficiency credentials with measurement data, not estimates.
Colocation billing and audit requirements. Singapore’s colocation market — Equinix SG1–SG5, ST Telemedia Global DC (SGP1, SGP3), Keppel Data Centres — charges power separately from space. Accurate per-tenant power billing requires outlet-level metering at ±0.5% or better. Disputes over PDU-reported consumption versus invoice amounts are a recurring colocation friction point when metering accuracy is poor. Intelligent PDUs with certified metering accuracy reduce that friction and provide the audit trail operators need for MAS or contractual compliance.
AI infrastructure density. Singapore’s data centre sector is undergoing rapid AI densification. AISG (AI Singapore) and the national AI strategy are driving GPU cluster deployments that were not part of facility planning five years ago. A 42U rack with eight NVIDIA H100 SXM5 servers draws up to 30 kW at full load — and spikes to 90+ kW at boot due to inrush. Standard metered PDUs cannot capture this. Without intelligent PDU data, capacity planning for AI racks defaults to conservative nameplate margins that strand significant available power. The Sunbird Auto Power Budget feature, documented in a Comcast deployment, recovered 40% additional available rack capacity after replacing estimated allocations with measured data — the same outcome is achievable in Singapore AI deployments.
Singapore context: A 1MW data centre hall running 4 kW per rack at Tier 2 metering (±3% accuracy) has a measurement uncertainty of ±30 kW — enough to mask capacity for 7–8 additional racks. The same hall with Tier 4 metering at ±0.5% reduces uncertainty to ±5 kW. At Singapore colocation rates, the difference between found and stranded capacity at 100-rack scale is material.
The Raritan PX4: what makes it the intelligent PDU specification for Singapore
The Raritan PX4 is the flagship intelligent rack PDU from Raritan (a Legrand brand) and is the product Enova supplies and supports in Singapore. Understanding what it does differently from baseline PDUs is useful before comparing it against alternatives.
Raritan PX4: key capabilities
- Metering accuracy: ±0.5% — the most accurate metering in the rack PDU category, suitable for colocation billing, Green Mark reporting, and capacity planning.
- Per-outlet power quality measurement — current (A), voltage (V), active power (W), apparent power (VA), power factor, total harmonic distortion (THD), reactive power (VAR), and accumulated energy (kWh) at each outlet individually.
- Inrush current capture — records peak inrush current at each outlet at power-on events. Critical for AI GPU deployments where boot-time inrush determines actual breaker and PDU sizing requirements.
- Residual Current Monitoring (RCM) — detects insulation leakage before it becomes a fault. Required in some Singapore critical facility frameworks and increasingly specified in financial institution data centres.
- HDOT high-density outlet design — removes unnecessary outlet moulding to fit more outlets in the PDU without increasing form factor.
- Xerus Technology Platform with iX Controller — the network intelligence module handling connectivity, firmware, alerting, and multi-protocol support (SNMP, REST API, Modbus TCP, DCIM integration).
- DCIM integration — native integration with Sunbird Power IQ and dcTrack, enabling real-time power chain visualisation, Auto Power Budget, and automated capacity reporting.
- Environmental sensor ports — supports Raritan temperature, humidity, differential air pressure, and contact closure sensors for combined thermal and power monitoring at the rack.
- Three-phase models available — for high-density and AI GPU rack deployments requiring 380–415V three-phase input at 16A, 32A, or higher.
Enova is an authorised Raritan distributor in Singapore. We stock PX4 models across the outlet-metered, outlet-metered switched, and intelligent ranges for single-phase and three-phase deployments.
Request a Raritan PX4 recommendation for your deployment →A practical checklist before you specify
Before selecting a PDU model, confirm the following with your facilities team and colo operator (if applicable):
[1] Feed type and current rating. Confirm whether your rack feed is single-phase or three-phase, the voltage (typically 230V single-phase or 400V three-phase in Singapore), and the rated current per feed (16A, 32A, 63A). This determines the PDU input specification entirely.
[2] Inlet connector type. Singapore colo facilities typically use IEC 309 (CEE) connectors for three-phase feeds and C20 for single-phase. US-spec (NEMA) connectors are common in US-origin equipment but may not match your facility’s feeds.
[3] Device outlet requirements. List every device going into the rack with its outlet type (C13 or C19) and power draw per PSU. Dual-PSU servers require two outlets each. GPU servers typically require C19 outlets.
[4] Peak versus nameplate draw. For AI or GPU server deployments, obtain the inrush or peak current specification from the server vendor. Raritan PX4 can measure this in situ on first boot — but the PDU and upstream breaker must be rated to handle it before that measurement is taken.
[5] DCIM integration requirement. If you are deploying Sunbird or another DCIM platform, confirm the PDU’s supported protocols before purchase. Raritan PX4 integrates natively with Sunbird Power IQ. Other DCIM platforms may require different PDU models or protocol configurations.
[6] Regulatory and compliance requirements. If the deployment is in an MAS-regulated financial institution, a government CII environment, or a facility subject to BCA Green Mark DC assessment, confirm whether RCM, metering accuracy certification, or specific audit trail capabilities are required.
What to ask for in a PDU quote
When requesting a quote for intelligent PDUs in Singapore, the following information allows a distributor to return an accurate recommendation rather than a range of options:
Rack count and outlet count per rack. Input voltage and phase per rack (confirm with facilities). Inlet connector type. Outlet type mix (C13/C19 ratio). PDU metering requirement (input only, outlet-level, or intelligent). Switching requirement (yes/no). RCM requirement (yes/no). DCIM platform in use. Mounting orientation (vertical or horizontal). Any colo-specific approved PDU list.
With those parameters, a specific Raritan PX4 model can be identified rather than a generic category. Enova handles Raritan PDU supply and commissioning support for Singapore and regional deployments.
Send your PDU requirements to Enova →Frequently asked questions
What is an intelligent PDU and how does it differ from a basic PDU?
A basic PDU is a power strip on a rail — it distributes power to rack equipment but provides no visibility into consumption. An intelligent PDU adds per-outlet power metering, remote outlet switching, environmental sensor ports, and network connectivity. The Raritan PX4 measures current, voltage, power factor, total harmonic distortion, and inrush current at the individual outlet level, with ±0.5% metering accuracy. That data feeds DCIM systems and enables capacity planning, anomaly detection, and compliance reporting.
What PDU tier do I need for a Singapore data centre Green Mark application?
BCA Green Mark for Data Centres requires demonstrable PUE measurement and energy monitoring across the facility. At the rack level, this means outlet-level metering at a minimum — you need to know how much power each server is drawing, not just the aggregate at the circuit breaker. Singapore’s DC-CFA2 requirements (PUE ≤1.25) and Green Mark DC assessments both favour facilities that can produce granular power data. Switched outlet-metered PDUs such as the Raritan PX4 provide the measurement depth required and integrate directly with DCIM platforms for automated reporting.
Do AI GPU racks require a different PDU specification?
Yes. AI GPU servers draw 3 to 5 times their rated load for 50 to 200 milliseconds at every power-on event. This inrush current causes breaker trips and PDU overload events that are invisible on standard metered PDUs, which only log average consumption. Intelligent PDUs with inrush current capture — such as the Raritan PX4 — record the actual peak current at each boot event. For racks at 15–30 kW or above, three-phase intelligent PDUs are the correct specification, and per-outlet metering is essential to understand real utilisation versus nameplate draw.
What is the difference between outlet-metered and outlet-metered switched PDU?
An outlet-metered PDU measures power at each outlet but cannot remotely control it. An outlet-metered switched PDU adds the ability to remotely power-cycle or shut off individual outlets via the network interface. For most data centre applications the switched capability is worth including: it allows remote server restart without a site visit, enables graceful shutdown of non-critical equipment during a power event, and provides the ability to lock out outlets to prevent unauthorised equipment connections.
What is RCM (Residual Current Monitoring) on a PDU and when is it required?
Residual Current Monitoring detects leakage current between live conductors and earth — the electrical signature of insulation degradation before it becomes a fault. RCM-equipped PDUs like the Raritan PX4 alert operators to developing insulation problems weeks or months before they cause a trip or fire hazard. RCM is required by some Singapore fire safety and electrical installation standards for critical facilities and is increasingly specified in colocation contracts and government data centre frameworks. It is especially relevant for high-density deployments where cable runs and connector counts are high.
Can I integrate rack PDU data with Sunbird DCIM?
Yes. Raritan PX4 PDUs integrate natively with Sunbird DCIM via the Sunbird dcTrack and Power IQ platforms. Power IQ ingests real-time per-outlet metering data from Raritan PDUs to produce power chain maps, capacity utilisation reports, and automated alerts. This integration enables the Auto Power Budget feature — where operators replaced estimated power allocations with measured data and discovered additional available rack capacity. Enova supplies both Raritan PDUs and Sunbird DCIM and can scope a joint deployment for Singapore environments.
What Raritan PDU models does Enova stock in Singapore?
Enova Technologies is an authorised Raritan distributor in Singapore and stocks the PX4 intelligent rack PDU range, including outlet-metered, outlet-metered switched, and high-density three-phase models. Contact [email protected] with your outlet count, input voltage, phase configuration, and rack density requirements for a specific model recommendation.
How many outlets does my rack PDU need?
The outlet count should match the device count in the rack plus a 20% buffer for growth. Most 1U/2U servers use one C13 outlet per power supply, with two outlets per server for dual-PSU configurations. High-density equipment typically uses C19 outlets at 16A or 20A. For AI GPU racks, per-server power draw is high enough that outlet count is secondary to per-outlet power rating — verify that the PDU’s outlet current rating matches the server’s actual draw, not nameplate only.
