What each IBM Power generation asks of the room, what it actually draws, and how much work it returns per watt. Every figure is IBM's own, charted across Power 5 through Power 11.
What the machine asks of the room, and what it gives back. Almost nobody puts those two side by side, because they live in different IBM documents: the electrical rating sits in the server specifications, the measured draw sits in IBM's ecodesign filings, and the performance figure sits in the Performance Capabilities Reference. This page charts all three together.
One finding runs through everything below. The rating is not the draw. Across Power 9 the measured maximum runs between 22 and 55 percent of the rated maximum, so a room built to the nameplate is usually built two to four times larger than it needs to be. That gap narrows sharply on Power 11, which changes a habit a lot of shops still rely on.
Every number here is IBM's own, and each section links out to the full per-machine-type reference for that generation.
The entry machine collapsed
The smallest IBM i server in each generation got heavier for twenty years and then fell off a cliff. A Power 9 S914 is rated at 1,600 W. The Power 11 S1112 that replaces it is rated at 550 W, a 66 percent drop in two generations, and it runs on a standard 120 V circuit.
This is the single most useful chart for a small shop, because it inverts the usual assumption. Moving forward two generations does not mean rewiring the room. It usually means the electrical work disappears from the project.
Rated maximum power, entry model of each generation
The Power 8 and Power 9 entry machines are 4U systems carrying a full scale-out chassis. The Power 11 S1112 is a half-wide 2U, and also ships as a tower, so a site with no rack at all can run one.
Source: IBM server specifications, per machine type
Rated against measured, and why the gap is closing
IBM publishes two different power figures for the same machine. The rated maximum is what the machine is permitted to pull, and it is what an electrician and an inspector work from. The measured maximum comes from IBM's filings under EU Regulation 2019/424, where IBM tests a high-end configuration of each machine type and reports what it actually drew.
The percentage on each bar is the measured maximum as a share of the rating. The trend is the story: Power 9 machines sit between 22 and 55 percent, while Power 11 machines sit between 68 and 86 percent.
That matters more than it looks. The old rule of thumb, that a Power server will never come close to its nameplate, was true for Power 9 and is no longer safe on Power 11. IBM rates the current generation much closer to what it will really do.
IBM measured maximum draw against rated maximum
Rated maximumIBM measured maximum
Measured figures are IBM's tested high-end configuration for each machine type, which is the worst case IBM files rather than a typical build. A lightly configured machine sits lower again. Size the circuit to the rating, because that is what code requires, and size the UPS and cooling budget to the measurement plus a real margin.
Source: IBM server specifications and IBM Technical Documentation for EU Regulation 2019/424
This is the comparison that actually decides a purchase, and it does not exist in any single IBM document. Divide the IBM i CPW rating of a machine by its rated maximum power and you get the work returned per watt of facilities capacity you had to provision.
The result reorders the Power 11 line completely. A Power S1122 is rated at 2,600 W and a Power S1124 at 2,950 W, a difference of 350 W or 13.5 percent, which is nothing in facilities terms. Yet a 60-core S1124 returns 456 CPW per rated watt against the S1122's 91. That is five times the IBM i output for a rounding error in electrical load.
If you are sizing a room for IBM i, this chart is the argument for buying the 4U machine.
IBM i CPW per rated watt, Power 11 whole-machine configurations
Only whole-machine CPW figures are charted. IBM publishes most S1122 processor features as a 4-core partition figure rather than a machine figure, because IBM i is capped at four cores per partition on those features, and charting a partition number beside a machine number would misstate both. The E1180 is per system node, and its figures scale linearly with node count, so the per-watt result is the same for a four-node system.
Source: IBM i 7.6 Performance Capabilities Reference (July 2026) for CPW, IBM server specifications for rated power
Within a single machine type, adding cores costs almost nothing in facilities terms. The rating does not move at all, because it is a property of the chassis and its power supplies rather than the processor feature installed in it.
An S1124 is rated at 2,950 W whether it holds 16 cores or 60. The room does not change. The output triples.
The cost of scaling up inside a generation is not electrical. It is the IBM i software tier, which steps from P20 to P30 between the 32-core and 48-core features and reprices the operating system, the licensed program products, and most third-party IBM i software for as long as you own the machine. Price the tier before you settle the processor feature.
Power S1124 (9824-42A): same room, three times the work
Processor feature
Cores
IBM i CPW
Rated power
CPW per watt
IBM i tier
EP3X
16
433,200
2,950 W
146.8
P20
EP3X
32
823,000
2,950 W
279.0
P20
EP3H
48
1,118,500
2,950 W
379.2
P30
EP3Y
60
1,345,900
2,950 W
456.2
P30
Rated power, thermal output, weight and rack space are identical across all four rows. Only the processor feature and the software tier change. The step from 32 to 48 cores is where the tier moves from P20 to P30, and that recurring software cost usually outweighs the one-time hardware difference.
The multi-drawer legacy machines are where a generation move pays for itself in facilities terms alone. A 32-core Power 6 Power 570 is four stacked processor drawers. A Power 11 S1124 replaces it in a single 4U chassis.
Read the last column. That is capacity handed back to the room, before any performance argument is made.
Where the room gets nothing back, the rating rises slightly while the work done inside it rises far more, so the comparison to make is the output per watt chart above rather than the rating alone. The two rows worth acting on are the multi-drawer Power 6 570 and the Power 9 S914, where the facilities saving is large and immediate.
How to use these numbers without getting them wrong
Four rules keep a plan honest, and each one is a mistake this page has seen made.
Size the circuit to the rating and the UPS to the measurement. Codes and inspectors work from nameplate; battery runtime is calculated against real load. Using one number for both jobs is wrong in one direction or the other.
Never compare a partition CPW figure to a machine CPW figure. IBM publishes some features as a 4-core partition rating because IBM i is capped there, and a naive comparison makes a capable machine look weak.
Multiply the drawers before anything else. Power 5 and Power 6 enterprise machines publish per enclosure, and a four-drawer system is four times every published figure, including weight and rack space.
Confirm the machine type, not the model name. A model name like S924 or Power 720 covers several machine types with different ratings, and on Power 8 the power supply feature decides what voltage the machine accepts.
Every AS400 and IBM Power machine type indexed by generation, with links to the full reference on each.
Working out what your room can take?
Send the machine type you run now and the one you are considering, and we will come back with the rated and measured figures for both, the circuit and cooling difference, and what the room gets back.