Rack PDU Selection for High-Density GPU Servers

Choosing the right rack power distribution unit (PDU) for high-density GPU servers prevents overloaded circuits, unexpected shutdowns, and costly downtime. This guide walks you through calculating your power budget, selecting the correct plug and receptacle types, and deciding between switched and monitored PDUs, so you can match the PDU to your actual hardware and facility constraints.

What You’ll Need

Server nameplate or spec sheet power ratings (max and typical)
Number of GPU servers per rack
Facility voltage and amperage available (e.g., 208V, 240V, 415V; 30A, 60A)
Knowledge of your rack’s physical layout (vertical mounting space, zero U vs. 1U)
List of receptacle types your servers use (C19, C13, or proprietary)
Optional: power monitoring software or a basic spreadsheet

Step-by-Step

Calculate your per-rack power budget. Add the maximum rated power (in watts) of every GPU server and network switch you plan to install in the rack. For example, a server with 4x 450W GPUs plus a 300W CPU and other components might draw 2,200W at peak. Multiply by the number of servers. Add a 20% safety margin for transient spikes and future expansion. This total is your minimum PDU capacity.
Determine your facility’s input power. Check the voltage and amperage your data center can supply to each rack. Common options are 208V single-phase, 208V three-phase, 240V single-phase, or 415V three-phase. The PDU input must match this exactly. If you have 208V three-phase, you can use a 3-phase PDU that balances loads across phases, which is more efficient than single-phase.
Convert your power budget to amps. Use the formula: Amps = Watts / (Volts × Power Factor). For most GPU servers, the power factor is close to 0.95–1.0. For example, 20,000W at 208V and PF 0.95 gives 20,000 / (208 × 0.95) ≈ 101A. This tells you the minimum input amperage. If you have three-phase, divide by √3 (1.732) for line-to-line voltage: Amps = Watts / (Volts × 1.732 × PF).
Choose the PDU input plug. Match the PDU’s input plug to your facility’s receptacle. Common types: NEMA L6-30 (30A, 208V), NEMA L15-30 (30A, 3-phase), IEC 60309 (60A, 3-phase), or hardwired. If you’re unsure, check with your facility team. Never use an adapter to change voltage or amperage—this is a safety risk.
Select the output receptacle types and count. Most high-density GPU servers use C19 outlets for the power supply (up to 16A each) and C13 for lower-power devices. Count how many C19 and C13 connections you need. A single server may have two C19 inlets for redundancy. Ensure the PDU has enough outlets of each type, plus a few spares.
Decide between switched, monitored, or basic PDU. A basic PDU just distributes power. A monitored PDU shows current, voltage, and power per outlet or per phase. A switched PDU allows remote on/off per outlet, which is useful for rebooting hung servers. For high-density GPU racks, at least monitored is recommended to track load and avoid overloads; switched is ideal for remote management.
Check physical form factor. Vertical (zero U) PDUs mount on the rear or side rails of the rack, saving rack space. Horizontal (1U) PDUs fit in the front or rear. For high-density racks, vertical is common because they can be longer and provide more outlets. Measure your rack’s internal height and clearance to ensure the PDU fits without blocking airflow.
Verify phase balancing (if three-phase). If you choose a 3-phase PDU, distribute your servers across the three phases. Each phase has a current limit. For example, a 60A 3-phase PDU has 60A per phase, not 180A total. Calculate the load per phase by grouping servers and ensure no phase exceeds its limit.
Consider redundant power feeds. If your servers have dual power supplies, you can connect each supply to a different PDU (A and B feeds). This protects against a single PDU failure. Ensure both PDUs are on separate circuit breakers and ideally separate UPS systems.
Plan for cable management. High-density servers have thick power cords. Use cable arms or vertical cable managers to keep cords organized and avoid blocking airflow. Ensure the PDU outlets are oriented so cords can be plugged without excessive bending.
Order and test. Before full deployment, install one PDU and plug in a few servers. Measure the actual current draw with a clamp meter or the PDU’s monitoring. Verify the load is within 80% of the PDU’s rated capacity. This leaves headroom for surges and prevents nuisance tripping.

Common Mistakes

Using nameplate power instead of actual draw. Nameplates list maximum theoretical power, which is often higher than real usage. Use actual measured or typical draw for more accurate sizing, but still keep a margin.
Ignoring phase balancing. On a 3-phase PDU, putting all servers on one phase overloads that phase while others are underused.
Mismatching plug types. Assuming your facility has a certain receptacle without checking. Always verify the exact plug and voltage.
Forgetting about inrush current. GPU servers can draw 2-3x their steady-state current for a few milliseconds at power-on. Your PDU and circuit breaker must handle this. Use a PDU with a higher rating or stagger server power-on.
Overloading outlets. A C19 outlet is rated for 16A, but the PDU’s total amperage is the limit. You can’t plug two 16A devices into two C19 outlets on a 30A PDU without exceeding the total.
Not planning for future expansion. Buying a PDU that exactly meets today’s needs leaves no room for additional servers or higher-power GPUs.

Tips

Use a PDU with a higher input rating than your calculated need (e.g., 60A instead of 30A) if space and budget allow. It provides flexibility and reduces the risk of overload.
Choose a PDU with a visible digital display for quick load checks without needing to log in to software.
For three-phase, prefer PDUs that allow per-phase monitoring so you can see imbalances.
If your servers support it, enable power capping in the BIOS or management controller to limit peak draw.
Label each PDU outlet with the server name and power supply number to simplify troubleshooting.
Consider using a PDU with a longer input cable (e.g., 3m or 5m) to reach floor or overhead power drops without extension cords.

Troubleshooting

PDU trips its breaker immediately after plugging in servers.
This usually means the total load exceeds the PDU’s rating or there’s a short circuit. Unplug all devices, reset the breaker, then plug in one server at a time while monitoring the current. If it trips with a single server, the server may have a faulty power supply.

Monitored PDU shows high load on one phase only.
Re-balance the servers across phases. Move some servers to different outlets that are on other phases. Check the PDU’s phase labeling (typically L1, L2, L3).

Server won’t power on, but PDU shows power.
Check the outlet’s circuit breaker (some PDUs have per-outlet breakers). Also verify the outlet is switched on if it’s a switched PDU. Try a different outlet to rule out a faulty receptacle.

Voltage reads low on the PDU display.
Low voltage (e.g., below 200V on a 208V circuit) can cause server instability. Check the facility’s transformer taps or upstream breaker. Contact your data center facility team—this is not a PDU problem.

Cannot connect to the PDU’s management interface.
Check the network cable and IP address. Reset the PDU to factory defaults if necessary. Ensure your management VLAN is reachable.

PDU gets hot to the touch.
Warmth is normal, but excessive heat indicates overload or poor airflow. Check the load and ensure the PDU is not coiled or blocked. If it’s above 60°C, reduce load or improve cooling.

Frequently Asked Questions

Q: Can I use a 30A PDU for a rack that draws 35A?
A: No. The PDU and its breaker will trip. You need at least a 40A or 60A PDU, and your facility must provide that circuit.

Q: What’s the difference between C13 and C19 outlets?
A: C13 is a smaller connector rated up to 10A (or 15A in some regions), commonly used for monitors and lower-power devices. C19 is larger and rated up to 16A, used for high-power servers and GPUs.

Q: Do I need a switched PDU for GPU servers?
A: Not strictly, but it’s highly recommended. Switched PDUs let you remotely reboot a hung server, which is common with high-density compute. Monitored-only PDUs are cheaper but don’t allow remote power cycling.

Q: How do I know if my facility has three-phase power?
A: Check your data center’s specifications or ask the facility manager. Three-phase receptacles have different plug shapes (e.g., L15-30 or IEC 60309) and typically provide 208V or 415V. Single-phase is usually 120V or 208V.

Q: Can I plug a 240V server into a 208V PDU?
A: Most server power supplies accept a range (e.g., 100-240V), so it will work, but the server may draw more amps at 208V than at 240V to achieve the same wattage. Check your server’s specs.

Q: What is the 80% rule?
A: It’s a safety guideline from the National Electrical Code (NEC) that continuous loads should not exceed 80% of the circuit breaker’s rating. For a 30A circuit, that’s 24A continuous. This prevents nuisance trips and overheating.

General steps for current mainstream versions; menus can differ by device and version. See the disclaimer or report an error.

More in this bed

The whole Tech How-To bed →