IBM Spyre Accelerator for Power
A 75 watt PCIe card that adds more than 300 TOPS of AI inference to a Power 11 server. It is the only IBM AI chip an AS/400 shop can actually order as an add-on, and it comes with a short list of things it refuses to work without.
Think of the Spyre Accelerator the way you would think of a specialist tool in a workshop. It is very fast at one job and completely useless without the bench, the power supply, and the operator who knows how to switch it on. IBM sells the tool. The bench and the operator are separate line items.
Spyre is a PCIe card that runs AI inference. Inference is the part where a trained model answers a question, as opposed to the part where a model gets built in the first place. IBM designed it so that inference can happen next to your own production data instead of leaving the building.
What IBM publishes about the card
Each of the 32 cores carries 2 MB of its own scratchpad memory, and the card supports INT4, INT8, FP8, and FP16 number formats. The lower formats are how a large model gets squeezed into 128 GB without falling over.
Four things it will not run without
An ENZ0 PCIe4 expansion drawer
The card does not go in the server body. IBM specifies the ENZ0 drawer, which is its own purchase, its own rack space, and its own power and cabling.
A Linux partition running Red Hat Enterprise Linux
IBM lists RHEL 9.6, 9.8, and 10.2. That means a partition beside IBM i, not inside it.
Red Hat AI Inference Server or Red Hat OpenShift AI
IBM names these two products by name as the supported way to drive the card. This is a licence, not a download.
A model to serve
The card is an engine with no cargo until you load a model onto it. IBM's own open Granite family is the obvious starting point.
The honest summary for an AS/400 shop: the card is the cheap part of the project. Budget the drawer, the Linux partition, the Red Hat subscription, and the people who will keep all four current. Confirm every number and requirement against IBM's current documentation before you sign anything, because support matrices move.
How it compares with what you already own
Spyre card against the acceleration already inside a Power 11 processor
| On-chip Matrix Math Acceleration | Spyre Accelerator | |
|---|---|---|
| Where it lives | Inside every Power 10 and Power 11 core | PCIe card in an ENZ0 expansion drawer |
| Extra hardware to buy | None, you already have it | Card plus drawer |
| Extra software required | Depends on the runtime you choose | Red Hat AI Inference Server or Red Hat OpenShift AI |
| Runs beside IBM i | Yes, same processor | Yes, in a separate Linux partition |
| Best fit | Smaller models, scoring and classification, first pilots | Larger generative and agentic models, steady production load |
Both sit on the same physical server. The question is never which is better, it is which one your workload actually needs.
Related
Power 11 Matrix Math Acceleration
The AI acceleration most AS/400 shops already own and have never switched on. It is inside every core of the processor, it costs nothing extra, and it is the cheapest honest answer to a first AI pilot.
IBM Artificial Intelligence Unit (AIU)
The research chip that became the card you can order. The AIU is why the Spyre Accelerator has 32 cores and why those cores look so much like the AI engine already inside an IBM Z processor.
Red Hat AI Inference Server
One of the two products IBM names as required before a Spyre card will do anything. It is the driver, the traffic controller, and the reason a 128 GB card can serve more than one person at a time.
IBM Granite Models
IBM's own family of open language models, and the cargo an accelerator card is built to carry. You can download them free, run them on your own hardware, and nobody meters the tokens.
Related Reading
The Spyre Card Will Not Start Without Red Hat. Here Is the Whole Stack.
IBM's own documentation lists Red Hat AI Inference Server or Red Hat OpenShift AI as a requirement for running a Spyre Accelerator on Power 11. Most budgets we see have the card and nothing else.
NorthPole, the AIU, and Where Your Spyre Card Came From
The card you can order for a Power 11 server has a family tree, and reading it tells you what IBM thinks AI hardware is for. Four chips, one stubborn argument: keep the model next to the data.