The IBM Processor Value Unit tables are the published schedules that assign a number of Processor Value Units (PVUs) to each processor core, according to the processor vendor, brand, type, model number and the maximum number of sockets on the server. IBM uses the Processor Value Unit as the unit of measure for much of its distributed middleware sold through Passport Advantage: the number of PVU entitlements a customer needs is the PVU rating of the processor technology multiplied by the number of processor cores made available to the program.[1] IBM describes the tables as the official source of PVU-per-core requirements for servers.[2]
IBM’s informational pages note that the exact terms for any program are set by the International Program License Agreement, the program’s License Information document and the agreement under which it was acquired; the tables and FAQs are educational material that sits under those documents.[3]
History
IBM introduced PVUs in 2006. According to IBM, the change was intended “to better reflect the relative value a customer can receive” from different processor technologies, and was prompted by two developments: the spread of multi-core chips, with several processor cores on one piece of silicon, and the growing use of server virtualization.[2] Throughout, IBM has kept the definition of a processor for PVU purposes as the individual processor core on a chip, so that a dual-core chip counts as two processors.[1]
When IBM places a new processor technology in the structure, it assesses relative performance using industry benchmarks, which may include transaction-processing benchmarks such as TPC-C and processor benchmarks such as SPECint and SPECjbb. IBM also names market conditions and the aim of keeping the structure simple as factors.[2] The tables are updated as new processor technologies become available, and each section carries the date on which it was published. The tables list only technologies that were generally available on that date, and IBM states that requirements for future processor technologies may differ.[1]
Structure of the tables
The tables are published in two sections.
Section 1, RISC and System z, has requirements as of 25 September 2024. It covers IBM POWER, IBM Z and LinuxONE, HP Itanium and PA-RISC, and Oracle, Sun and Fujitsu SPARC processors. For POWER servers the rating depends on the server family: in the POWER11, POWER10, POWER9 and POWER8 generations the largest enterprise servers (for example the E1180 and E1080) are rated at 120 PVUs per core, mid-range four-socket models (E1150, E1050, E950, E850) at 100, and scale-out models at 70. Any POWER core running Linux, and any Integrated Facility for Linux (IFL), is rated at 70. The larger IBM Z machines from the z10 EC through the z17 ME1 are rated at 120 PVUs per core and the smaller models at 100, with each IFL or Central Processor engine counted as one core.[1]
Section 2, x86, has requirements as of 08 May 2024. Intel Xeon processors launched after the Nehalem generation (November 2008), including Xeon 6 and 5th-generation Xeon Scalable, are rated at 70, 100 or 120 PVUs per core for servers with a maximum of two, four, or more than four sockets. Pre-Nehalem Xeon models are rated at 50. AMD EPYC is rated at 70 and AMD Opteron at 50. Intel Core i3, i5, i7 and i9 are rated at 70, although IBM excludes the newest generation of Intel Core processors that use the Performance Hybrid Architecture with two different core types. Any x86 processor not otherwise listed is rated at 100.[1]
IBM characterizes a processor technology by five attributes: vendor, brand, type (for example dual-core or quad-core), the maximum number of sockets possible on the server, and, where the table requires it, the processor model number.[2]
Socket count and SMP servers
For processors whose rating depends on socket count, the relevant figure is the maximum number of sockets the server can hold, not the number populated. IBM justifies the socket buckets as a measure of server scalability, arguing that higher scalability allows higher software utilization and consolidation.[2] If sockets on two or more servers are connected to form a symmetric multiprocessing (SMP) server, the maximum socket count of the combined server applies. IBM’s example connects two 2-socket servers with 6 cores per socket; the result is treated as a 4-socket server with 24 cores, requiring 2,400 PVUs (100 per core times 24).[1]
Counting
The required entitlement for a server is the PVU-per-core rating multiplied by the total number of processor cores. IBM’s worked examples include a 2-socket server with 6 cores per socket (840 PVUs, at 70 per core times 12 cores), a 4-socket server with 6 cores per socket (2,400 PVUs) and an 8-socket server with 6 cores per socket (5,760 PVUs).[1] For POWER the FAQ gives an S822 with two 8-core sockets at 1,120 PVUs, an E850 with four 8-core sockets at 3,200 PVUs, and a Power 880 with eight 8-core sockets at 7,680 PVUs.[2]
| Server | Max sockets | Cores | PVU per core | PVUs required |
|---|---|---|---|---|
| x86 Xeon, 2 x 6 cores | 2 | 12 | 70 | 840 |
| x86 Xeon, 4 x 6 cores | 4 | 24 | 100 | 2,400 |
| x86 Xeon, 8 x 6 cores | >4 | 48 | 120 | 5,760 |
| POWER S822, 2 x 8 cores | 2 | 16 | 70 | 1,120 |
| POWER E850, 4 x 8 cores | 4 | 32 | 100 | 3,200 |
| POWER 880, 8 x 8 cores | 8 | 64 | 120 | 7,680 |
Activated, deactivated and inactive cores
Licences are needed for activated processor cores only. IBM defines an activated core as one “available for use in a physical or virtual server”, whether or not its capacity is limited by virtualization, operating-system commands or BIOS settings.[1] Cores shipped deactivated for future growth need no entitlement until they are turned on, and additional licences become due at the moment more cores are activated (rule).[4] A similar logic applies to POWER7 TurboCore mode, in which four of eight cores remain active after an IPL: IBM states that the inactive cores run no workload and need no entitlements, but all eight must be licensed if the server returns to normal mode (rule).[2]
Client devices
Laptops, desktops and workstations follow a separate rule rather than the server tables. For AMD and Intel x86 client devices IBM specifies 100 PVUs per core for single-core technology and 50 PVUs per core for multi-core technology, except multi-core Intel Core i3, i5 and i7, which are 70 PVUs per core (rule).[2]
Minimum quantities and transfers
A few PVU programs are priced so low that one PVU would cost less than one US dollar; for those part numbers IBM set a minimum order of 10 PVUs “for ease of ordering, billing, and administration” (rule).[2] Separately, the minimum requirement for any core-based licence is the entitlement for one whole processor core.[4]
PVUs for the same program are fully transferable among servers within an enterprise. When a program moves to a server whose cores carry a different rating, more or fewer PVUs may be needed, and any shortfall must be acquired before deployment (rule).[2]
Virtualization and tooling
Under full-capacity licensing the table rating applies to every activated core in the physical server. Under Virtualization Capacity (sub-capacity) licensing the same rating is applied to the cores made available to the program, subject to the counting rules described in IBM sub-capacity licensing and ILMT.[1]
Because the rating for many processors depends on socket count, the IBM License Metric Tool needs hypervisor data to choose the right row. When no connection to the VM manager is defined, ILMT applies default sub-capacity counting and assigns the highest PVU-per-core value listed for that processor; in IBM’s example a two-socket Xeon host that should be rated at 70 is charged at 120 per core, and virtual cores are not capped at physical capacity. IBM states that it accepts audit reports calculated this way rather than treating the client as ineligible for sub-capacity (rule).[5]
Some programs, Db2 among them, grant instances running in warm standby a fixed consumption instead of a table-based count. ILMT reports these with the PVU Warm Standby metric at a fixed 100 PVUs per endpoint, which is added to the program’s sub-capacity total.[6]
Containers
Programs deployed under IBM’s Container Licensing policy are not counted from the processor-model tables. For container licensing IBM states that “1 VPC is equal to 1 vCPU and 70 PVUs is equal to 1 vCPU”, with vCPU capacity derived from Kubernetes CPU limits and IBM License Service as the only accepted tracking tool.[7] That conversion is specific to containers and is not used for virtual machines; see IBM Passport Advantage licensing for the container counting method.
Out of scope
- Program prices per PVU, which vary by program and by Passport Advantage volume level.
- The Processor Model Number Discovery Guide and the online PVU calculator, which IBM offers as aids alongside the tables.
- Mainframe z/OS metrics such as MSUs, which are not governed by the PVU tables.
- A complete reprint of every row of the tables; readers should consult the current table for the exact processor model.