Server Power & PDU Buyer's Guide

PSU Sizing — Connected Load and the 60-80% Sweet Spot

PSU sizing starts with total connected load at sustained peak. Add up nameplate wattage for CPUs (at TDP), GPUs (at sustained, not peak), drives (5-10W per spinning drive, 3-5W per SSD), fans (10-20W per fan), and add-in cards. Idle wattage is irrelevant — peak sustained is what matters.

Once you have peak sustained load, size the PSU at 60-80% of nameplate rating. Modern Platinum and Titanium PSUs hit peak efficiency at 50% load and degrade above 80%. A 1100W PSU running 700W (64% load) delivers 94-95% efficiency and runs cool. The same PSU running 1000W (91% load) drops to 92% efficiency, runs hot, and degrades 2-3x faster.

For redundant N+1 PSU configurations, size each PSU to handle full load alone. Two 1100W PSUs in N+1 redundancy means each PSU should be sized at 60-80% of 1100W = 660-880W maximum draw. Both PSUs share load in normal operation; in failover, the surviving PSU takes 100% — which must still be within rating.

Redundancy — N+1, 2N, and the Rack PDU Side

N+1 redundancy means one extra PSU above what the chassis needs. For a server requiring 800W, N+1 is two 800W PSUs (or two 1100W in a more conservative spec). Single-PSU failure does not cause shutdown.

2N redundancy is full duplication — every PSU has a partner on a separate power feed. 2N is standard for tier-3 and tier-4 data centres with dual A-side / B-side feed.

For racks with dual-feed PDUs, the chassis-side PSU redundancy matters less than the PDU-side. A single rack PDU failure should not exceed the partner PDU's headroom. Size PDUs at 80% of breaker rating sustained — never 100%. A 30A PDU breaker can deliver 30A peak but should never sustain above 24A.

Common mistake: undersizing the PSU side correctly but oversizing the PDU side. Result: two PSUs running cool, one PDU running hot.

Hot-Swap Behaviour and BMC Health

Server-grade PSUs are typically hot-swap. Release latch, slide out the failed PSU, slide in the replacement, done — no chassis power-cycle, no VM disruption. Confirm "Hot Swap" in the PSU datasheet before assuming.

Workstation-grade or smaller chassis sometimes uses non-hot-swap PSUs. These require full chassis shutdown to replace.

BMC reporting on PSU health is the operational gate. iDRAC (Dell), iLO (HPE), IMM/XCC (Lenovo), and IPMI (Supermicro) all report PSU status. Healthy PSUs report "Present + OK"; degraded PSUs report fan-speed anomalies, output-voltage drift, or temperature warnings. Alert on any state other than "Present + OK" in monitoring.

UPS Sizing and Runtime

UPS sizing has two dimensions: capacity (watts/VA) and runtime (minutes at rated load). Capacity matches the connected load plus headroom; runtime determines how long the rack survives a feed failure before generator pickup or graceful shutdown.

For commercial offices with generator backup, 5-15 minutes of UPS runtime is sufficient — generator picks up within 30 seconds and the UPS bridges the gap.

For data centres without generator backup, 30-60 minutes of runtime is typical — enough time to safely shut down or migrate workloads.

For edge / branch sites, runtime depends on power-grid reliability. Sites with frequent brownouts may want 30+ minutes; sites with stable grid power need only 5-10 minutes.

Online double-conversion UPS (true sine wave output, always converting) is the recommended topology for any production workload. Line-interactive UPS is acceptable for office endpoints but produces stepped-approximation output that some modern PSUs reject under battery load.

Predictive Failure and Battery Replacement Cycles

PSU failure indicators visible from BMC: fan-speed anomalies (PSU fan ramping when chassis load is unchanged), output-voltage drift beyond ±5% of nominal, temperature warnings during normal load. Any of these warrants ordering a replacement PSU within the next maintenance window.

UPS battery replacement is typically required every 3-5 years. Symptoms of aging batteries include runtime degradation (capacity drops below 80% of rated), self-test failures, and visible swelling on individual cells. Most enterprise UPS units expose battery health via SNMP — alert on remaining capacity below 75%.

Pro Disk Network ships UPS Replacement Battery Cartridges (RBCs) for the major APC, CyberPower, Tripp Lite, and Eaton families. Hot-swap battery replacement keeps the protected load powered during the swap. Document the battery replacement on your maintenance calendar at install time — most UPS failures are predictable battery-end-of-life events, not sudden component failures.