Capacity on Demand (CoD)
Capacity on Demand is IBM's way to switch on processor cores and memory units that are already installed in a Power server but sitting inactive. For an IBM i shop it is how you give a new AI inference partition cores and memory without a hardware install and, for a permanent activation, without a server restart.
What does Capacity on Demand do?
IBM ships many Power servers with more processor cores and more memory inside the box than you paid to use. Capacity on Demand, usually written CoD, is the set of offerings that turn the extra hardware on. IBM's own words are that you can activate inactive processor cores or memory units that are already installed on your server, on a temporary and permanent basis.
Think of a house wired for ten lights where only four switches work. The wire for the other six is already in the wall. CoD is the key that makes those six switches live. You are not paying an electrician to run new wire. You are paying to flip a switch.
This matters for AI work because an inference partition needs cores and memory that your production IBM i partition is already using. CoD is how a shop finds that capacity inside a server it already owns, instead of buying a second machine.
Which Capacity on Demand offerings does Power 11 document?
The six offerings IBM lists under Capacity on Demand for Power 11
| Offering | What it activates | How long it lasts |
|---|---|---|
| Capacity Upgrade on Demand (CUoD) | One or more inactive processor cores or memory units | Permanent, with no server restart and no interruption to your business |
| Trial Capacity on Demand | Inactive processor cores, memory, or both, at no charge | 30 power-on days, and the clock only advances while the server is powered on |
| Elastic Capacity on Demand | Processor cores and memory units, on and then off again | A number of days you specify, pre-paid, and changeable while the request is running |
| Power Enterprise Pool | Mobile CoD processor and memory resources shared by a group of systems | Moves between the systems in the pool |
| IBM Power Private Cloud with Shared Utility Capacity | Base capacity plus the usage above it | Ongoing, and IBM's product page describes metering by the minute of RAM and CPU |
| PowerVM Editions | The virtualization feature itself, not capacity | Permanent, activated with a code the same way capacity is |
Offering names and wording from IBM's Capacity on Demand pages in the Power 11 documentation for the Power E1150 (9043-MRU) and the Power S1122 (9824-22A), which list the same six. IBM calls Shared Utility Capacity Power Enterprise Pools 2.0, and called Elastic CoD On and Off CoD before. The per-minute metering line is from IBM's Flexible Capacity on Demand product page, which is marketing material rather than hardware documentation. IBM also documents advanced functions beside these, such as Enterprise Enablement and Active Memory Expansion.
How do you give an inference partition more cores?
Four stops between inactive hardware and a running partition
- Check what is already in the boxCompare how many cores and how much memory the server holds against how many are active today. That gap is what CoD can reach. If the hardware is not installed, CoD cannot help and you are buying parts.
- Pick permanent or temporaryCapacity Upgrade on Demand is permanent. Elastic CoD runs for a set number of days and is pre-paid. Trial CoD is free for 30 power-on days, which is the safe way to see whether the extra cores help your model before you pay.
- Get the code and enter itIBM says that after you place your order you receive a code that activates the inactive processor cores or memory units. The code is tied to one server, by machine type and serial number.
- Hand the new capacity to the partitionActivation makes the resource available to the server. Dynamic logical partitioning is what moves it into a running partition. Nothing lands in your inference partition until you assign it.
IBM states that you must use the Hardware Management Console to manage Trial Capacity on Demand activations, and that you must use it to manage Elastic Capacity on Demand as well. Sequence drawn from IBM's Power 11 Capacity on Demand documentation.
Can you switch on a Spyre card the same way?
No. Every resource IBM names in its Capacity on Demand documentation is a processor core or a memory unit. A Spyre Accelerator for Power is a PCIe Gen5 card, feature code ECSE, that someone installs in a slot. No activation code makes a card appear in an empty slot.
- Cores and memory: often already inside the server, so CoD can switch them on
- Spyre Accelerator cards: a physical install in a PCIe slot, so order, ship and fit
- Memory in a Power 11 server rides on DDIMM modules, so an activation still needs the modules seated
Plan the two separately. Cores and memory for an inference partition can often come from capacity you already own. Accelerator cards cannot. IBM's Flexible Capacity on Demand product page does not list accelerators or GPUs among the resources you can activate.
Related Terms
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.
Hardware Management Console
The HMC is the console an IBM i shop uses to create the partition an AI model runs in, then hand that partition processors, memory and the PCIe slot holding an accelerator card while production keeps running. IBM Power 11 servers need HMC release V11R1M1110 or later.
Differential DIMM (DDIMM)
A DDIMM is the memory card an IBM Power 11 or Power 10 server takes in place of a plain DDR memory stick. It carries its own buffer chip and plugs into an Open Memory Interface slot, which is how a two-socket Power S1124 reaches the 2456 GBps memory bandwidth ceiling IBM lists for feeding data into an AI model.
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.