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.
What is it?
The Power E1180 is the biggest server in IBM's Power 11 line. IBM calls it machine type 9080-HEU. You do not buy it as one sealed box. You buy it in drawers. Each drawer is a 5U unit holding four processor modules and its own memory, and one small 2U control unit sits with them to run the whole thing. Think of a chest of drawers you add to over time. Start with one drawer and add up to four as the work grows. Cores and memory climb with every drawer you add.
What IBM lists for a full Power E1180
| Item | What IBM lists |
|---|---|
| Machine type model | 9080-HEU |
| Drawers | Up to four 5U processing drawers plus one 2U System Control Unit |
| Cores | Up to 256, with eight threads per core |
| Processor modules | 10, 12 or 16 core Single-Chip Modules at up to 4.40 GHz |
| Memory | Up to 64 TB DDR5, 16 TB per drawer |
| Memory bandwidth | Up to 512 GBps per socket, 2,048 GBps per drawer |
| PCIe slots in the body | Eight PCIe Gen5 slots per drawer, 32 across four |
| IBM i minimum levels | 7.4 TR12, 7.5 TR6, or 7.6 with the Power11 GA PTF Group |
Figures are IBM's own, from IBM Redbooks publication SG24-8587, IBM Power E1180 Technical Overview and Introduction, published 26 September 2025. SMT8 is IBM's name for running eight threads at once on one core. GBps means gigabytes per second.
Why does it matter for AI?
IBM gives Power 11 two kinds of AI help. The first is built into every core. It is called Matrix Math Acceleration, and it speeds up the math a model does without any extra card. The second is a card you add, the Spyre Accelerator, aimed at generative AI and larger model inference. The E1180 matters because it holds the most of those cards. IBM's Spyre publication lists 8 to 48 cards per system on 9080-HEU. Every other Power 11 model in that same table stops at 24 or at 12. If the plan is to serve several models at once, in production, on the machine where the data already lives, this is the only Power model IBM takes past 24 cards.
- On-chip Matrix Math Acceleration comes with the cores and needs no card
- The Spyre Accelerator is a separate card for generative AI and larger model inference
- Only the 9080-HEU is listed for more than 24 Spyre cards
How do 48 cards fit in one server?
They do not go in the server body. The body has eight PCIe Gen5 slots per drawer, which is 32 across a full four drawer machine, and those slots are wanted for storage and network cards too. The Spyre cards go in a separate box instead. It is the ENZ0 PCIe Gen4 expansion drawer, and each processing drawer supports up to four of them. Each expansion drawer holds up to 12 Gen4 adapter slots. That is where the room for 48 cards comes from.
The three things that add up to 48 cards
- Start with the slots in the bodyEach processing drawer has eight PCIe Gen5 slots. A full four drawer machine has 32. Storage and network cards want some of those.
- Add expansion drawersEach processing drawer supports up to four ENZ0 PCIe Gen4 expansion drawers. Each one holds up to 12 Gen4 adapter slots.
- Order Spyre in groupsIBM lists Spyre on 9080-HEU as 8 to 48 cards per system, in groups of 8 or 12. A 12 card group needs Red Hat AI Inference Server 3.5 or later.
One logical partition still takes only 24 cards. IBM says a system with more than 24 cards installed must use more than one partition.
PCIe Gen5 slots in the Power E1180 body
Spyre cards do not use these slots. They go in ENZ0 PCIe Gen4 expansion drawers, up to four per processing drawer, each holding up to 12 adapter slots.
Can IBM i use the Spyre cards?
Not directly. IBM's Spyre stack for Power runs on Red Hat Enterprise Linux 9.6, 9.8 or 10.2, with Podman and Quadlet as the container runtime. IBM i is not on that list. What you get instead is both jobs on one machine. IBM i runs in its own partition. The Linux inference partition runs beside it and owns the cards. The E1180 supports IBM i 7.4 TR12, 7.5 TR6, or 7.6 with the Power11 GA PTF Group, either native or as a client of Virtual I/O Server. So your data does not have to leave the building to get an answer from a model.
IBM lists up to 24 Spyre cards per logical partition. If more than 24 cards are installed in the machine, IBM says you must split them across more than one partition. Cards can be moved between running partitions with dynamic logical partitioning.
What power and cooling does IBM list?
A machine this size is a facilities question as much as a computing one. IBM publishes an electrical characteristics table for the 9080-HEU. Read the numbers below as published maximums, not as what a given order will draw.
Electrical characteristics IBM publishes for the Power E1180
| Property | IBM figure |
|---|---|
| Operating voltage | 200 to 208 or 220 to 240 V AC |
| Operating frequency | 50 Hz or 60 Hz |
| Phase | Single |
| Maximum power consumption | 4850 W |
| Maximum power source loading | 5 kVA |
| Maximum thermal output | 16548 BTU per hour |
These are IBM's published maximums for the configuration its specifications table covers, not a quote for a specific order. IBM points to the IBM Systems Energy Estimator for that. IBM also requires the rack to have a rear door and side panels to meet electromagnetic compatibility rules.
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.
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.
Dynamic Logical Partitioning
Dynamic logical partitioning, or DLPAR, moves processors, memory and card slots in and out of a running IBM Power partition without restarting it. IBM documents it for the Spyre Accelerator, so an AI card can join or leave a Linux inference partition while the IBM i production partition keeps working.
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.