Identifying Server Power Supplies: Matching and Sizing
Power supplies are ordered on wattage and they fail on everything else. A unit with the correct output that does not match the chassis, the other supply, or the input voltage is not a replacement.
They are not generic
The starting point, because it is where the assumption goes wrong.
Server power supplies are specific to the chassis family and generation. The physical form, the connector that mates with the backplane, the firmware, and the management interface all differ between platforms — and between generations of the same platform.
Two things follow.
Wattage is not the specification. Two 800W supplies from the same vendor may be for different chassis families and not interchangeable.
The part number encodes the rest. Which is why the reliable approach is the same as with drives: read the label on a fitted unit. Our part numbers guide covers why option and spare numbers can differ for the same item.
Match supplies to each other
The rule most often broken when adding a second supply for redundancy.
In a redundant pair, both supplies should be the same model and wattage. Mixing them causes problems that vary by platform: the system may refuse the mismatch, may run without redundancy, may report a persistent warning, or may limit both to the lower rating.
Where a chassis genuinely supports mixed wattage, it is documented as such rather than assumed. If in doubt, match.
The same applies to efficiency ratings in some platforms, which flag mismatched efficiency classes even where the wattage matches.
Input voltage changes the output
The specification most often missed, and it catches organisations buying internationally.
Many enterprise power supplies deliver their full rated wattage only on higher input voltage, and meaningfully less on standard lower-voltage supply.
So a supply labelled at a given wattage may not deliver it in your building. Two consequences.
Check the derated figure for your input voltage, not the headline number, when sizing.
A configuration that worked in one region may not in another — our importing guide covers this alongside connector differences.
This matters most on densely populated chassis and anything carrying accelerators, where the margin is smallest.
Sizing, and the redundancy trap
Covered fully in our sizing guide, and the essential point belongs here.
In a redundant configuration, each supply must carry the entire load on its own, because when one fails the other takes everything. Sizing two supplies to share a load they cannot individually carry means losing one takes the server down — which is the opposite of what redundancy was bought for.
And note that summing nameplate ratings across a rack produces a figure two or three times too high. Use measured draw where you have it, as our PDU guide covers.
Hot-plug and how supplies actually fail
Two practical points.
Most enterprise supplies are hot-plug, meaning a failed unit can be replaced while the server runs — provided the other one is working and the load is within what it can carry alone.
Fans are frequently what fail, rather than the power electronics. A supply whose fan is failing may report a fault, run hot, or become noticeably loud before it stops. That noise is a warning worth acting on.
Also worth knowing: a supply reporting a fault while still delivering power is a redundancy failure rather than an outage, and it should be treated as urgent rather than logged and forgotten. The system is now running on one.
Identifying what you have
Three methods.
Query the management controller. It reports installed supplies including model, wattage, status and frequently input voltage and current draw, without opening anything. Our management guide covers access.
Read the label on a fitted unit — visible from the rear on most rack servers without removing anything. Photograph it.
Check both supplies, not just one, since a mismatched pair may already be present and explain a warning nobody investigated.
Record the part number in your asset register. On end-of-life platforms this is the second most valuable thing to know after drives, since power supplies are the second most common failure.
What to hold as a spare
For any server whose failure matters, one supply per chassis type.
The reasoning is the same as for drives. Supplies fail, they are hot-plug on most platforms, and a spare on the shelf converts a redundancy failure into a five-minute swap rather than a wait. They cost the same bought now or in an emergency.
Where several servers share a chassis family, one spare covers all of them — which is another argument for standardising on fewer platforms.
What to send with an enquiry
The part number from a fitted supply, or a photograph of its label. The server model and generation. How many supplies are fitted and how many the chassis takes. Your input voltage. And whether you are replacing one or adding a second for redundancy, since the second case means matching what is already there.
Common questions
Can I use any supply of the right wattage?
No. Supplies are specific to the chassis family and generation — physical form, backplane connector, firmware and management interface all differ. Two 800W supplies from the same vendor may not be interchangeable.
Can I mix wattages in a redundant pair?
Generally no. Depending on the platform the system may refuse the mismatch, run without redundancy, warn persistently, or limit both to the lower rating. Where mixing is supported it is documented rather than assumed.
Will my supply deliver its rated wattage?
Not necessarily. Many enterprise supplies deliver full rated output only on higher input voltage and meaningfully less on standard lower-voltage supply. Size against the derated figure for your input, not the headline number.
What usually fails in a power supply?
Frequently the fan rather than the power electronics. A supply whose fan is failing may report a fault, run hot, or become noticeably loud before it stops — and that noise is a warning worth acting on.
Is a faulted supply urgent if the server is still running?
Yes. The system is now running on one supply with no redundancy, so a second event takes it down. Treat it as urgent rather than logging it and moving on.
Send us a photograph of the label on a fitted supply and your server model, and we will confirm the match.
