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.
Built intoEvery Power 10 and Power 11 core
JobSpeeds up AI inferencing math
Needs a GPUNo
Works forIBM i, AIX and Linux
Why it matters for an IBM i shop
You can run AI inferencing on the same server that already runs your production partitions. There is no GPU or extra card to buy, per IBM's published Power 11 AI Redbook guidance.
Data types it handles
- FP32
- BFloat16
- INT8
How it compares with the Spyre card
Matrix Math Acceleration vs the Spyre card
| Matrix Math Acceleration | IBM Spyre Accelerator | |
|---|---|---|
| What it is | Built into every processor core | A separate PCIe4 card |
| Where you find it | Every Power 10 and Power 11 core | Up to 16 cards in a Power 11 expansion drawer |
| Usable from IBM i | Yes, along with AIX and Linux | Not documented as of September 2026 |