Server Total Cost of Ownership: Counting What It Actually Costs
Two servers with the same specification can cost very different amounts to run, and over a few years the difference frequently exceeds the purchase price. This covers counting what a server actually costs.
The costs beyond the purchase
Five, and most organisations count one.
Electricity for the server itself. A server drawing a few hundred watts continuously costs a meaningful amount per year, and that is before cooling.
Electricity for cooling. Every watt the server consumes becomes heat that must be removed, and removing it costs more electricity. Depending on the cooling arrangement, this can approach the server’s own draw.
Rack space. In a colocation facility this is billed directly. In your own room it is real but invisible, which is why it gets ignored.
Support and maintenance. Contracts, spares holdings, and the time people spend.
Licensing. Frequently the largest line, and frequently tied to cores or sockets rather than to the machine. Our support guide covers what varies.
Where older hardware costs more
Being honest, because this cuts against buying used.
Older generations do the same work for more power. A machine two or three generations old can consume substantially more electricity for equivalent output, and that difference runs every hour for years.
Two situations where it decides the purchase.
Continuous high-load operation. A machine running near capacity around the clock accumulates the difference fastest.
Expensive electricity or constrained cooling. Where power costs are high or the room is at its cooling limit, efficiency stops being a background cost and becomes the constraint.
Our used server guide covers when used is right, and this is the main argument on the other side.
Where older hardware still wins
Three situations, and they are common.
Intermittent or light load. A machine idling most of the time consumes far less than its rating, and the efficiency difference between generations shrinks accordingly.
Short remaining horizon. A machine needed for two years accumulates two years of difference, not eight.
Where the purchase saving is large. The power difference has to exceed a substantial purchase gap, and frequently it does not within the intended life.
The calculation is simple: annual power difference times years of intended use, against the purchase difference. Do it with real numbers rather than assuming either way.
Licensing: the line that changes the answer
Where licensing is per core or per socket, the hardware decision becomes a licensing decision.
Two effects.
More cores can cost more in licence than in hardware. A processor upgrade that doubles cores can double a licence bill that already exceeds the server.
Fewer, faster cores are sometimes cheaper overall than more, slower ones, even at higher hardware cost, because the licence follows the core count.
Check the licensing model before choosing processors, not after. Our second processor guide covers where this bites hardest.
Consolidation changes the arithmetic
The lever most organisations have and do not pull.
Several lightly-used machines cost several machines’ worth of power, cooling, rack space and support. Consolidating them onto fewer, better-utilised hosts reduces every one of those lines at once.
Two caveats. Consolidation concentrates risk, so redundancy matters more afterwards. And licensing may or may not improve, depending on the model. Our consolidation guide covers planning it.
A practical way to decide
Four steps, using real figures.
Measure actual draw rather than using the supply rating. A machine with a 750W supply may draw 200W. Management controllers report actual consumption on most enterprise servers.
Multiply by your electricity cost for a year, then add a cooling multiplier appropriate to your arrangement.
Add licensing, support and rack cost for the same year.
Multiply by the intended life and compare options on that total rather than on purchase price.
Frequently the answer is not the one you expected, in either direction. Our lifecycle guide covers using this to time refreshes.
Frequently asked questions
What costs should I count beyond the purchase price?
Electricity for the server, electricity for cooling it, rack space, support and spares, and licensing. Licensing is frequently the largest line and is often tied to cores rather than the machine.
Does used hardware cost more to run?
Older generations do the same work for more power, and that difference runs every hour. It matters most under continuous high load, where electricity is expensive, or where the room is at its cooling limit.
When is older hardware still cheaper overall?
Under intermittent or light load, where the machine is needed for a short horizon, and where the purchase saving is large. The power difference has to exceed the purchase gap within the intended life, and frequently it does not.
How do I measure how much power a server uses?
From the management controller, which reports actual consumption on most enterprise servers. Do not use the power supply rating — a machine with a 750W supply may draw 200W.
Can more cores cost more in licensing than hardware?
Frequently. Where licensing is per core, a processor upgrade that doubles cores can double a licence bill that already exceeds the server cost. Sometimes fewer faster cores are cheaper overall despite higher hardware cost.
How much does cooling add?
Every watt the server consumes becomes heat that must be removed, and removing it costs more electricity. Depending on the cooling arrangement, this can approach the server’s own draw.
Tell us the machines you are comparing and your electricity cost, and we will help you cost them over the life rather than at purchase.
