Glossary

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.

What it isHypervisor in firmware
Guests it runsAIX, IBM i, Linux
Smallest core slice0.05 of a core
Edition todayEnterprise, included

What does PowerVM do for an IBM i shop?

A Power server is one box. PowerVM is the part that splits it into several smaller computers. Each one is called a logical partition, or LPAR. Picture a building with one furnace and separate locked apartments. The furnace is the hardware. The apartments are the partitions. Nobody in one apartment can walk into another.

That split is what makes AI work on a production machine. IBM's Power 11 documentation says PowerVM runs AIX, IBM i and Linux virtual machines on the same server. So the production IBM i workload keeps its own partition, and a Red Hat Enterprise Linux partition beside it runs the AI model. You can patch, restart or delete the Linux side without touching IBM i.

How small a slice of a server can one partition get?

Dedicated and shared processor partitions under PowerVM

How the partition gets CPUDedicated processorShared processor
Smallest amount of CPU1 whole core0.05 of a core
Step size when you change it1 whole core0.01 of a core
How the hypervisor runs itA whole core is mapped to the partitionTime sliced in a 10 millisecond dispatch window
What caps the partition countThe number of activated coresThe server model, up to 1000 virtual machines on some

Source: Introduction to IBM PowerVM, IBM Redbook SG24-8535, March 2023.

What are the parts of PowerVM?

  • PowerVM hypervisor, shortened to PHYP. It sits in firmware and hands out processors, memory and card slots to each partition.
  • Virtual I/O Server, or VIOS. It is a partition itself. It owns real network and disk cards and shares them with the other partitions, so each one does not need its own.
  • Service processor. IBM lists this as an enterprise Baseboard Management Controller, or eBMC, or a Flexible Service Processor, or FSP.
  • Hardware Management Console, or HMC. The console a person uses to create and change partitions.
  • PowerVM NovaLink and IBM Power Virtualization Center, or PowerVC. Two ways to drive the same work from software instead of by hand.

What does an AI partition need on top of PowerVM?

A Spyre Accelerator card is not shared between partitions. IBM assigns it to one partition, and one partition can hold up to 24 cards. That partition has to run Red Hat Enterprise Linux 9.6, 9.8 or 10.2 with Red Hat AI Inference Server. IBM lists a floor of 512 GB of memory, 10 processor cores and 600 GB of storage per model you serve, and says to raise the memory as the model size and pod count grow.

Spyre Accelerator cards, per partition and per system

One logical partition24 cards
Power E1180 system8 to 48 cards

Source: IBM Spyre Accelerator for Power, September 2026. The per-system range is for machine type 9080-HEU and comes in groups of 8 or 12.

The card is not an IBM i device. IBM drives it from a Red Hat Enterprise Linux partition on the same physical server, so an IBM i shop can try AI inference on hardware it already owns without moving the production partition.

Related Terms