Shared Processor Pool
A group of Power cores that several partitions draw from, with one ceiling over the whole group. It is how an IBM i shop gives a Linux AI partition real processor capacity on the production server while holding both that partition and the IBM i license count under a fixed limit.
What is a shared processor pool?
A Power server is cut into logical partitions, or LPARs. A partition can own whole cores, or it can own a fraction of a core. IBM calls the fractional kind a micropartition. Every micropartition takes its processor time from a shared processor pool, or SPP, which is simply a group of physical cores with one meter on it.
Think of a water tank with several taps. Each tap is promised a minimum flow. The tank has a limit nobody can go past. Every core you did not hand to a dedicated partition sits in the default pool, which IBM creates for you. You can also build your own pool and set the ceiling yourself.
How small can a partition be on Power 11?
Shared processor limits on a Power 11 server
| Limit | Power 11 value |
|---|---|
| Smallest capacity one partition can hold | 0.05 processing units, which is 0.05 of a physical core |
| Partitions that can share one physical core | Up to 20 |
| Virtual processors a partition can define | Up to 20 for each processing unit assigned to it |
| Pools on one server | One default pool, up to 64 in total |
| What one processing unit equals | The processing capacity of one physical processor |
Source: IBM Power11 Scale-Out Servers: Introduction and Overview, IBM Redbook SG24-8590, first edition October 2025. It applies to models 9824-42A, 9856-42H, 9824-22A and 9856-22H.
The PowerVM hypervisor, which IBM shortens to PHYP, is what hands out that time. It dispatches and time slices the shared partitions across the real cores. A person creates and changes the partitions from the Hardware Management Console, or HMC.
What happens when the AI partition gets busy?
Each partition is set either capped or uncapped. A capped partition stops at its entitlement no matter how much of the pool is idle. An uncapped partition may borrow idle capacity from its own pool. IBM's Power 11 Redbook says that under heavy load every micropartition is still assured its entitlement, plus whatever it draws from the pool reserve if it is uncapped, and that capacity nobody uses is ceded to the uncapped partitions in the same pool according to their weighting.
- Entitlement is the amount a partition is guaranteed. It is the floor, not the ceiling.
- An uncapped partition can pass its entitlement, but it can never exceed its virtual processor count.
- A dedicated partition can be set to donate its idle cycles to a pool and still keep absolute priority over its own cores.
- The pool maximum caps the whole group, so a busy AI partition cannot drain capacity you meant to keep for production.
IBM's PowerVM Redbook SG24-8535, published March 2023, puts the uncapped weighting scale at 0 to 255 and says the reserved processing units of a pool default to zero. Those two settings decide who gets spare capacity first, so check them before you assume an AI partition can burst.
Why does the pool decide your IBM i license count?
How IBM counts processors for an IBM i per-processor license, in three steps
- Count the whole machineAdd up every activated processor in the server. That is the first number.
- Count the partitions insteadAdd the cores assigned to IBM i partitions that use dedicated processors. Then, for each shared processor pool, add the smaller of the pool maximum capacity or the sum of the virtual processors of each uncapped partition plus the processing units of each capped partition running IBM i. Round any leftover fraction up to a whole processor.
- Use whichever number is smallerIBM counts the smaller of step one and step two. That is why a pool ceiling can lower what you owe.
Source: IBM i license terms, IBM Redbook SG24-8590, October 2025. Confirm your own agreement with IBM before you rely on a count.
IBM's own example: three IBM i licenses on a four processor server
| Partition | Operating system | Mode | Processing units | Virtual processors | Most processors it can use |
|---|---|---|---|---|---|
| A | IBM i | Uncapped | 1.75 | 2 | 2.00 |
| B | IBM i | Capped | 0.60 | 1 | 0.60 |
| C | IBM i | Capped | 0.40 | 1 | 0.40 |
| D | Linux | Uncapped | 1.25 | 2 | 2.00 |
Source: IBM Power Systems Logical partitioning, IBM, copyright 2018 to 2026. All four partitions share one pool on a four processor machine. The three IBM i partitions can reach 3.00 processors together, which matches the three licenses, and the Linux partition shares the same machine. IBM's own text notes that switching B and C to uncapped breaks that compliance.
Where does a Spyre AI partition fit in a pool?
The Spyre Accelerator is driven from a Linux partition. IBM's Power 11 Redbook lists the minimum stack for general availability in the fourth quarter of 2025 as firmware FW1110.10, HMC level HMC1111, Red Hat Enterprise Linux 9.6, the Spyre software stack container, and Red Hat OpenShift AI, which was a technology preview in the fourth quarter of 2025 and generally available in the first quarter of 2026. Put that Linux partition in its own pool and you have set a hard ceiling on how much processor capacity your AI work can take from the box your business runs on.
A pool controls processor capacity only. It does not cap memory, and it does not assign accelerator cards. IBM says the Spyre Accelerator reads the server's main DDR5 memory rather than memory on the card, and that the cards live in the accelerator expansion drawer, up to eight per drawer. Size memory and card slots separately from the pool.
Related Terms
PowerVM
IBM's hypervisor for Power servers, built into system firmware. It is what lets a Linux partition running an AI model sit on the same machine as the production IBM i partition, each with its own processors, memory and cards.
AI Inference Partition
A separate supported operating-system partition used to execute an AI model while an IBM i partition may remain the production application and data source.
Dynamic Logical Partitioning
Dynamic logical partitioning, or DLPAR, moves processors, memory and card slots in and out of a running IBM Power partition without restarting it. IBM documents it for the Spyre Accelerator, so an AI card can join or leave a Linux inference partition while the IBM i production partition keeps working.
Hardware Management Console
The HMC is the console an IBM i shop uses to create the partition an AI model runs in, then hand that partition processors, memory and the PCIe slot holding an accelerator card while production keeps running. IBM Power 11 servers need HMC release V11R1M1110 or later.