Glossary
Matrix Math Acceleration (MMA)
An on-chip acceleration feature in IBM Power10 and Power11 processors that speeds up the matrix math operations used in AI inferencing, without requiring a separate GPU.
Where it livesInside every Power 10 and Power 11 processor core
Extra card to buyNone
IBM names it forLocal inference on the Power S1112, July 2026
Runs every model inside IBM iNo
What Matrix Math Acceleration is
Matrix math is the arithmetic an AI model does most. Matrix Math Acceleration, MMA for short, is a feature built into the Power 10 and Power 11 processor that does that arithmetic faster. It is on the chip. There is no card to order and no drawer to fill.
MMA versus the Spyre Accelerator card
Two ways IBM Power does AI math
| Matrix Math Acceleration | Spyre Accelerator | |
|---|---|---|
| Where it lives | Inside the processor core | A separate PCIe card |
| Hardware to add | None | The card and an ENZ0 expansion drawer |
| IBM's example | Local inference on the Power S1112 | Driven from a Linux partition on Power 11 |
What to check before an MMA pilot
Four checks IBM's S1112 example assumes
- Operating-system supportConfirm the inference runtime is supported on the partition you plan to use.
- Model frameworkConfirm the model's framework can use MMA.
- Cores and memoryCount what the runtime needs and what is actually free.
- Production headroomMake sure the IBM i partitions keep their reserve.
On-chip does not mean every model runs natively inside IBM i. The feature speeds up supported runtimes on the same physical system. Which partition runs the model is still a placement decision.