IBM Power S1122
The Power S1122 is the two-socket 2U IBM Power 11 server, machine type 9824-22A, and IBM lists it for 8 or 12 Spyre AI accelerator cards ... half the ceiling of the larger S1124. For an IBM i shop it is the smaller Power 11 route into on-site model inference, with one trap: only the ERGR processors let an IBM i partition grow past 4 cores.
What is it?
The Power S1122 is IBM's two-socket Power 11 server in a 2U box. IBM calls it machine type 9824-22A. It is the size below the 4U Power S1124. Think of the S1124 as a tall filing cabinet and the S1122 as the two-drawer version that fits under a desk. Same kind of work, less room inside. It runs IBM i, AIX and Linux, and it can run them side by side in separate partitions.
What IBM and its configuration data list for the Power S1122
| Item | What is listed |
|---|---|
| Machine type model | 9824-22A |
| Size and sockets | 2U, two processor sockets |
| General availability | 8 July 2025, announcement letter AD25-0035 |
| Cores | 4 to 60 Power 11 cores |
| Processor features | ERGR 4 or 8 core and ERGQ 10 or 20 core entry single-chip modules, EBG8 16 or 32 core, EBG9 24 or 48 core and EBGA 30 or 60 core dual-chip modules |
| Clock speeds | 3.6 to 4 GHz on ERGR, 3.05 to 4 GHz on ERGQ, 3 to 4.2 GHz on EBG8, 2.65 to 4.15 GHz on EBG9, 2.4 to 3.95 GHz on EBGA |
| Memory | 64 GB to 4 TB DDR5 across 32 slots |
| Internal storage | Up to 8 NVMe U.2 bays |
| PCIe slots in the body | Four Gen4 x16 or four Gen5 x8, four Gen5 x8 in x16 connectors, two Gen4 x8 |
| Operating systems | IBM i, AIX and Linux |
| IBM i minimum level | IBM i 7.4 TR12 or later |
The 2U size, the 60 core ceiling, the 4 TB memory ceiling and the operating system list are IBM's own figures from its Power S1122 product page. The availability date and announcement letter are from IBM's product lifecycle record for 9824-22A. Processor feature codes, clock speeds, memory slot count, bay count, slot mix and the IBM i level are from the Midland Information Systems configuration data for 9824-22A. An entry single-chip module (eSCM) puts one processor chip in a socket; a dual-chip module (DCM) puts two, which is why the core counts double. U.2 is the connector shape used by the internal flash drives.
How many Spyre cards does it take?
IBM's Spyre Accelerator for Power publication lists 9824-22A at 8 or 12 cards per system. You cannot order one card or three. You order a supported step, and the S1122 only gets two of them. Each card holds 32 accelerator cores and 128 GB of its own memory, so 8 cards give you 256 accelerator cores and 1 TB of AI memory, and 12 cards give you 384 cores and 1.5 TB. Everything about this is a Linux-side order, not an IBM i one.
AI accelerator cores at the top card step IBM lists for each Power 11 model
IBM lists 32 accelerator cores and 128 GB of Low Power Double Data Rate 5 (LPDDR5) memory on each PCIe5 x16 AI Accelerator adapter, feature code ECSE, CCIN 2E1F, built on 5 nm technology inside a 75 watt power envelope. IBM's Table 1 lists 8 or 12 cards on 9824-22A, 9824-22B and 9856-22H; 8, 12, 16 or 24 on 9043-MRU, 9824-42A and 9856-42H; and 8 to 48 in groups of 8 or 12 on 9080-HEU, the Power E1180. One logical partition holds up to 24 cards no matter which machine it runs on, so an S1122 never hits that ceiling.
What is the catch with 12 cards?
Power. IBM says a drawer holding 12 cards has to run those cards at reduced power, because of what the drawer can supply. A drawer holding 8 cards can run them at full power for peak performance. It works like 12 people sharing one power strip: everyone gets a plug, but nobody gets full wattage. The S1122 only has these two steps, so this is the whole decision on this model rather than a detail.
- Reduced power is the default from Red Hat AI Inference Server 3.5 onward, and a 12-card drawer needs that release or later
- Full power on an 8-card drawer is a manual setup: you set two environment variables when the container starts
- Changing those variables after the container has started does nothing, so the setting has to be right at start time
- In a 12-card drawer IBM puts 8 cards in x16 slots and 4 in x8 slots, and says to use the x16 cards when a model is split across four cards
IBM says the lower power limit can reduce performance compared with the high power mode available on an 8-card drawer, and that it generally hits retrieval and entity extraction work harder than it hits document embedding work. So 12 cards is not simply 50 percent more machine than 8 cards. Decide by the job you are running, not by the card count.
Can IBM i use the cards?
Not directly. IBM's Spyre software for Power runs on Red Hat Enterprise Linux, and IBM i is not on that list. What you get instead is both jobs on one machine. IBM i runs in its own partition at 7.4 TR12 or later. A Linux inference partition runs beside it and owns the cards. Your data never has to leave the building to get an answer from a model.
Three things to order before an S1122 can run Spyre
- An ENZ0 drawer with ENZA fanout modulesThe cards do not go in the server body. IBM says the Spyre card is supported only in ENZ0 PCIe Gen4 expansion drawers fitted with feature code ENZA accelerator fanout modules. The standard ENZF fanout modules are for ordinary input and output adapters and will not take the card.
- Room for the inference partitionIBM lists a floor of 512 GB of memory, 10 processor cores and 600 GB of storage for every model you serve. The S1122 tops out at 4 TB of memory and 60 cores, so the floor fits. But it is real space, and it comes out of the same machine your production work runs on.
- Red Hat software on a Linux partitionIBM requires Red Hat AI Inference Server or Red Hat OpenShift AI on Red Hat Enterprise Linux 9.6, 9.8 or 10.2, with Podman and Quadlet as the container runtime. A drawer holding 12 cards needs Red Hat AI Inference Server 3.5 or later.
So a Spyre quote on an S1122 is never just a card quote. It is cards, plus a drawer with the right fanout modules, plus the Red Hat software, plus memory and cores set aside for the partition that drives the cards.
What does the processor choice cost IBM i?
More on this model than on any other Power 11 server, so read this before you pick a feature code. Midland Information Systems' configuration data says only the ERGR entry single-chip modules support running IBM i natively, virtualized, or both. On every other S1122 processor option, an IBM i partition is held to a 4-core maximum. That means a 60-core S1122 can still only give one IBM i partition what 4 cores can do.
Most work one IBM i partition can hold on a Power 11 S1122
CPW is Commercial Processing Workload, IBM's own rating for how much IBM i transaction work a machine can carry. These figures are from the Midland Information Systems configuration data for 9824-22A, which rates the 8-core ERGR at 236,400 CPW, or 29,550 CPW per core, and holds IBM i partitions on all other processor options to the 4-core maximum of 118,200 CPW. A larger core count on a dual-chip module buys AIX and Linux room, not IBM i room.
This is a partition sizing rule, not a Spyre rule. IBM's Spyre Accelerator for Power publication lists supported machine types and card counts and states no processor condition for 9824-22A. Treat the processor choice and the accelerator order as two separate questions, and confirm both with IBM or your reseller before you buy.
Related Terms
Spyre Accelerator
IBM's PCIe AI inference card. On IBM Power 11 it carries 32 cores, 128 GB of memory, and more than 300 TOPS inside a 75 watt envelope.
IBM Power S1124
The Power S1124 is the largest scale-out IBM Power 11 server, machine type 9824-42A, and IBM lists it for 8, 12, 16 or 24 Spyre AI accelerator cards ... but only when both processor modules are installed. An IBM i shop can keep production on IBM i 7.4 TR12 or later and run model inference in a Red Hat Enterprise Linux partition on the same 4U machine.
IBM Power E1180
The Power E1180 is the high-end IBM Power 11 server, machine type 9080-HEU, and the only Power model IBM lists for more than 24 Spyre AI accelerator cards ... up to 48 per system. An IBM i shop can keep production on IBM i 7.4 TR12 or later and run large AI inference in a Red Hat Enterprise Linux partition on the same machine.
ENZ0 Expansion Drawer
The PCIe4 expansion drawer IBM documents for supported Spyre Accelerator for Power configurations on Power 11 systems.
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.
Red Hat AI Inference Server
Red Hat's model serving product, built on the open source vLLM engine. IBM lists it as one of two supported ways to drive a Spyre Accelerator on IBM Power.
AI Workload Placement
The decision about which system, operating environment, and partition executes an AI model or agent, separately from the IBM i application supplying its data.