Processor Value Units on Intel, Power and mainframe.
The PVU formula looks identical on every platform, but the inputs are not. Intel, Power, and mainframe each carry their own per core ratings and their own counting rules, and the platform you run on can swing the bill before a single core is miscounted. Here is how the three differ where it matters.
PVU per core times allocated cores is the constant. What changes from platform to platform is the rating each core carries in the IBM Processor Value Unit table and the way cores are counted toward the IBM software. Treat all three platforms as if they followed Intel rules and you will misstate the position, almost always in the direction that favors IBM.
Intel
Most distributed IBM middleware runs on x86, and the table assigns these cores a moderate rating. As a working figure, a typical Intel core rates around 70 PVU, though the exact value depends on the processor model. On Intel the live questions are usually about virtualization: which hypervisor is in play, whether sub-capacity standing holds, and whether the report counted allocated virtual cores or quietly picked up the full physical host. The rating is rarely the surprise on Intel; the core count is.
Power
Power processors carry higher per core ratings than typical Intel cores, reflecting their throughput, so the same number of cores costs more PVU on Power than on x86. Power also counts differently. In an LPAR configuration the allocation of cores to a partition is what drives the count, and capacity assigned to a partition is counted even when it sits idle. That makes Power configuration decisions, and any uplift from on demand capacity features, directly material to the PVU result.
Mainframe
On the mainframe, PVU based products are sized against the engines and capacity the software is entitled to run on, and mainframe engines carry their own ratings in the table. Specialty engines and capacity settings complicate the picture, and the relevant question is which capacity the IBM software is actually licensed against rather than the gross capacity of the machine. Mainframe positions reward precision because a single rating or capacity assumption is multiplied across high value engines.
On Power LPAR, cores allocated to a partition count toward PVU even if they are never busy. Buyers who assume only active cores are charged routinely understate the position. The metric follows entitlement and allocation, not utilization.
Where the platform difference becomes a finding
- Wrong family rating. Applying an Intel style rating to Power or mainframe cores understates the position, while applying a high rating to the wrong cores overstates it. Either way the number is wrong.
- Utilization confused with allocation. Counting only busy cores on Power understates PVU and invites a correction in IBM's favor.
- Capacity features ignored. On demand capacity on Power can lift the cores that count, and reports that miss it leave the position exposed.
- Mixed estates averaged. Treating a mixed Intel, Power, and mainframe estate with one assumed rating produces a number that is wrong on every platform at once.
The PVU formula is constant but the platform is not. Intel, Power, and mainframe carry different ratings and different counting rules, and idle Power cores still count. Confirm the correct table rating and counting rule for each platform in scope before accepting any PVU total, because a single wrong assumption is multiplied across every core.