Difference between revisions of "POWER9"
|Line 38:||Line 38:|
[[PowerNV|PowerNV]] systems use [[SMT4|SMT4]]
[[PowerNV|PowerNV]] systems use [[SMT4|SMT4]] exclusively, and chipsare .
are planned to be made in both [[Scale Out|Scale Out]] (direct-attach RAM) and [[Scale Up|Scale Up]] (centaur-buffered RAM) configurations.<ref>Stuecheli, Jeff. POWER9. Presentation for [https://www.ibm.com/developerworks/community/wikis/home?lang=en#!/wiki/Power+Systems/page/AIX+Virtual+User+Group+-+USA AIX VUG]. ([https://public.dhe.ibm.com/systems/power/community/aix/Central-VUG-Replays/2017-01-26_IBM_POWER9.wmv video download], [https://www.ibm.com/developerworks/community/wikis/form/anonymous/api/wiki/61ad9cf2-c6a3-4d2c-b779-61ff0266d32a/page/1cb956e8-4160-4bea-a956-e51490c2b920/attachment/56cea2a9-a574-4fbb-8b2c-675432367250/media/POWER9-VUG.pdf slides], [[User:Torpcoms/Timemark/POWER9|timemarks]])</ref>
== Modules ==
== Modules ==
Revision as of 13:06, 22 February 2018
|Maximum cores||12 SMT8 / 24 SMT4|
|L2 cache / slice||512kB|
|L3 cache / slice||10MB|
|Production availability||January 2018|
|← POWER8E||POWER10 →|
POWER9 is IBM's most recent POWER compatible server and workstation CPU (POWER ISA v3.0B). Built on a 14nm process, each CPU package can contain up to 24 SMT4 cores or 12 SMT8 cores. Each pair of SMT4 cores, or singleton SMT8 core, comprises a slice; each slice in turn contains 512kB L2 cache and 10MB L3 cache. Raptor Computing Systems' 4- and 8-core processors provide unpaired cores, such that one SMT4 core per slice is fused off. This allows each of the SMT4 cores to utilize the full cache of the slice exclusively, increasing performance for these ST-focused processors.
|Scale Out||Nimbus||unknown[note 1]|
- The presentation by Jeff Stuecheli makes it clear that these chips will exist, but the codename for them is currently unknown.
Chips are planned to be made in both Scale Out (direct-attach RAM) and Scale Up (centaur-buffered RAM) configurations; where a Scale Out system can use normal RAM attached directly to the CPU, Scale Up chips require that access to RAM be through a Centaur memory buffer, which behaves like a L4 cache.
Nimbus chips are available in three different modules: Sforza, Monza, and LaGrange. Each module exposes different I/O functionality to the host platform, allowing purpose-built systems to be constructed in addition to more general-purpose computers. Sforza is the most flexible of these packages, providing PCIe 4.0 lanes as the main I/O resource, and is what Talos™ II uses for maximal similarity to existing desktop, workstation, and server systems. Monza modules offer the most OpenCAPI/NVLink bandwidth and are used in IBM's AC922 (Witherspoon) systems, such as those used by the Sierra and Summit supercomputers. LaGrange modules offer increased XBus bandwidth between processor sockets and are used by the Google/Rackspace Zaius motherboard used in the Barreleye G2 system.
Little is known about Cumulus chips now; as Scale Up chips, they will trade some peripherals bandwidth for communication between more than 2 sockets.