UPS Sizing for Servers: VA vs Watts, Runtime and Battery Replacement
A UPS is judged entirely on a day it's never been tested. Two mistakes cause most failures: sizing...
Power supplies are the component buyers think about last and regret first. They're not exciting, they don't appear on a spec comparison, and the instinct when configuring a server is simple: pick the biggest one available and stop worrying about it.
That instinct is wrong, and expensively so. A PSU running at 15% load is less efficient than a correctly sized unit running at 50% — meaning an oversized 1600W supply in a server drawing 300W wastes more energy, generates more heat, and costs more to run than an 800W unit doing the same job. In a climate like Qatar's, where every wasted watt becomes heat your air conditioning has to remove, that inefficiency gets paid for twice.
This guide covers how redundancy actually works, what the 80 PLUS ratings mean in practice, how to size a PSU correctly, and what to check when buying refurbished.
Redundancy: What N+1 and 2N Actually Mean
Nearly every enterprise server ships with two PSU bays, and most refurbished units arrive with both populated. The terminology describes how much failure the configuration can absorb:
| Configuration | What It Means | Survives |
|---|---|---|
| Single (N) | One PSU carrying the full load | Nothing — PSU failure means server down |
| N+1 redundant | One spare beyond the minimum needed; load shared across units | One PSU failure, with no interruption |
| 2N (fully redundant) | Two independent paths, each able to carry 100% of load | A full power path failure, including upstream |
For a typical dual-PSU server, N+1 is what you have: both supplies share the load, and if one fails the other picks up the full burden with no downtime. Hot-swap design means the failed unit can be replaced while the server keeps running.
Redundancy only counts if both PSUs are fed from genuinely separate sources. Two supplies plugged into the same power strip protects against a PSU failing — not against the strip, breaker, or UPS failing. Feed each PSU from a different circuit or UPS where possible.
80 PLUS Efficiency Ratings Explained
The 80 PLUS certification measures how much drawn power reaches the components rather than being lost as heat. A 90% efficient PSU delivering 500W to the server pulls roughly 555W from the wall — the missing 55W becomes heat in your server room.
| Tier | 20% Load | 50% Load | 100% Load |
|---|---|---|---|
| Bronze | 82% | 85% | 82% |
| Silver | 85% | 89% | 85% |
| Gold | 87% | 90% | 87% |
| Platinum | 90% | 92% | 89% |
| Titanium | 94% | 96% | 91% |
Two things stand out in that table. First, efficiency peaks at around 50% load on every tier — not at full load. Second, Titanium's real distinction is holding high efficiency down at low loads, where lower tiers fall away sharply.
For most business deployments, Platinum is the sensible target: it captures the large majority of the practical savings over Gold, without Titanium's price premium. Titanium earns its cost in high-density racks, in facilities paying for both power and the cooling to remove its heat, or where the server genuinely runs at low utilisation most of the time.
Sizing: Why Bigger Is Not Better
This is the mistake worth understanding properly, because it runs against instinct. Since efficiency peaks near 50% load, the aim is to choose a PSU capacity that places your typical load in the 40–60% band.
A worked example. A dual-CPU server with 256GB RAM and eight drives might draw around 400W under normal load:
- 800W PSU — 400W is 50% load, sitting exactly in the efficiency sweet spot
- 1600W PSU — the same 400W is 25% load, where efficiency is measurably lower, and the unit costs more upfront
The oversized option costs more to buy, more to run, and produces more waste heat. To size correctly:
- Start from actual component draw — CPU TDPs, populated DIMMs, drive count, and any add-in cards. Use vendor TDP figures rather than theoretical maximums.
- Add 20–30% headroom — for transient spikes and planned expansion, not as a blanket safety multiplier
- Account for what you'll actually add — GPUs are the big one. A GPU for AI inference can add 150–300W each, which changes the calculation entirely and typically requires 1100W+ supplies.
- Remember redundancy changes the maths — in an N+1 pair sharing load, each PSU carries roughly half the total, so both must individually be capable of the full load during a failure
Both iDRAC (Dell) and iLO (HPE) report real-time power draw. If you already own the server, read the actual consumption rather than estimating from a spec sheet — it's usually lower than people expect, which often means a smaller PSU is the better choice.
Input Voltage: The Detail That Changes Output
A specification that catches people out: many high-wattage server PSUs cannot deliver their rated output on low-voltage mains. A supply rated 2000W might deliver only around 1400W at 100–127V, reaching full output only at 200–240V.
Qatar's 240V supply works in your favour here — high-wattage PSUs deliver their full rated output without derating. Worth knowing if you're buying hardware originally specified for a 110V market, or reading a datasheet written for one.
Efficiency and Cooling: The Qatar Double Cost
Every watt lost to PSU inefficiency becomes heat inside the chassis, and every watt of heat inside the chassis has to be removed by the room's cooling. In a temperate climate that's a modest overhead. In Qatar, where air conditioning already works against 45°C summer ambient temperatures, you effectively pay for that wasted energy twice — once at the wall, once at the AC unit.
A practical illustration: a Bronze PSU at 82% efficiency delivering 500W wastes roughly 110W as heat. A Platinum unit at 92% wastes about 43W for the same delivered power. Across several servers running continuously, that difference is meaningful both on the electricity bill and on the cooling load your server room air conditioning has to handle.
When sizing cooling capacity, use the figure drawn from the wall rather than the server's delivered power. An inefficient PSU increases both numbers, which is why efficiency ratings matter more in hot climates than the headline energy saving suggests.
Buying PSUs for Refurbished Servers
Power supplies are among the most sensible components to buy refurbished — they're simple, robust devices with no firmware to age and a long service life. A few points specific to used hardware:
- PSUs are generation and chassis specific — Dell and HPE use their own form factors, and even within a brand, 13th-gen and 14th-gen PowerEdge supplies aren't interchangeable. Always match to the exact server model.
- Match wattage across both bays — servers generally expect matched pairs. Mixing a 495W and an 1100W supply typically triggers a warning or disables redundancy.
- Check the efficiency tier on the label — the same wattage was often sold in Gold and Platinum variants; the badge is printed on the unit
- Keep a spare on the shelf — PSUs are inexpensive relative to downtime, hot-swappable, and the single easiest failure to recover from if you have the part in the building rather than on order
At ServerDove Trading & Services, we stock tested power supplies for Dell PowerEdge and HPE ProLiant servers across generations, with local warranty and delivery in Doha and nationwide. Send us your server model and we'll confirm the correct part and wattage.
Frequently Asked Questions
Can I run a server with only one PSU installed?
Yes, provided that single supply is rated for the server's full load. The server will run normally but report a redundancy warning, and any PSU failure takes the machine offline. Acceptable for test or non-critical systems; not advisable for anything production.
Do both power supplies need to be the same wattage?
Yes, in practice. Enterprise servers expect matched pairs, and mismatched wattages usually trigger a warning or disable redundancy entirely. Always replace with the same wattage and, ideally, the same efficiency tier.
Is a bigger PSU always better?
No. PSU efficiency peaks around 50% load, so an oversized supply running at 15–25% load is less efficient than a correctly sized one at 50%. Oversizing costs more upfront, wastes more energy, and generates more heat. Size so typical load sits in the 40–60% range.
Is Titanium worth the extra cost over Platinum?
For most business deployments, Platinum captures the majority of practical savings without the premium. Titanium makes sense in high-density racks, where you pay for both power and the cooling to remove its heat, or where servers run at consistently low utilisation — Titanium holds efficiency far better at low loads.
Can I hot-swap a failed server power supply?
Yes, on any server with redundant PSUs. The remaining supply carries the full load while the failed unit is removed and replaced, with no downtime. This is one of the main practical benefits of redundant configurations.
Where can I buy server power supplies in Qatar?
ServerDove stocks tested power supplies for Dell PowerEdge and HPE ProLiant servers across multiple generations, with delivery in Doha and nationwide and local warranty support. Provide your server model and we'll confirm the correct part number and wattage.
Final Thoughts
Three decisions cover almost everything worth getting right: run redundant supplies fed from separate circuits, choose Platinum unless there's a specific reason for Titanium, and size so your typical load lands near 50% rather than reaching for the biggest unit on the list.
In Qatar the efficiency argument carries more weight than the raw electricity saving suggests, because wasted watts become heat your cooling system pays to remove. A correctly sized Platinum PSU is quietly one of the better value decisions in a server build.
Need a replacement PSU or advice on sizing? Send us your server model and configuration and we'll confirm the right part — with delivery across Qatar and local warranty support. Request a Quote · Chat on WhatsApp
