Editing InfiniBand (section)

== Specification ==
Specifications are published by the InfiniBand trade association.

=== Performance ===
Original names for speeds were single-data rate (SDR), double-data rate (DDR) and quad-data rate (QDR) as given below.<ref name="comeback" /> Subsequently, other three-letter acronyms were added for even higher data rates.<ref name="fdr_fact_sheet">{{Cite web |date=November 11, 2021 |title=FDR InfiniBand Fact Sheet |url=https://cw.infinibandta.org/document/dl/7260 |access-date=September 30, 2021 |publisher=InfiniBand Trade Association |archive-date=August 26, 2016 |archive-url=https://web.archive.org/web/20160826064526/https://cw.infinibandta.org/document/dl/7260 |url-status=dead }}</ref>

{| class="wikitable"
|+ InfiniBand unidirectional data rates
! rowspan="2" |
! rowspan="2" |Year<ref name="ccgrid11-ib-hse-23">{{cite web |last=Panda |first=Dhabaleswar K. |author2=Sayantan Sur |date=2011 |title=Network Speed Acceleration with IB and HSE |url=http://www.ics.uci.edu/~ccgrid11/files/ccgrid11-ib-hse_last.pdf#page=23 |access-date=13 September 2014 |work=Designing Cloud and Grid Computing Systems with InfiniBand and High-Speed Ethernet |publisher=CCGrid 2011 |pages=23 |location=Newport Beach, CA, USA}}</ref>
! colspan="2" rowspan="2" |Line code
! rowspan="2" |Signaling rate (Gbit/s)
! colspan="4" |[[Throughput]] (Gbit/s)<ref name="ib_over">{{Cite web |title=InfiniBand Roadmap: IBTA - InfiniBand Trade Association |url=http://www.infinibandta.org/content/pages.php?pg=technology_overview |url-status=dead |archive-url=https://web.archive.org/web/20110929111021/http://www.infinibandta.org/content/pages.php?pg=technology_overview |archive-date=2011-09-29 |access-date=2009-10-27}}</ref>
! rowspan="2" |Adapter latency (μs)<ref>http://www.hpcadvisorycouncil.com/events/2014/swiss-workshop/presos/Day_1/1_Mellanox.pdf // Mellanox</ref>
|-
!1x
!4x
!8x
!12x
|-
!{{abbr|SDR|Single Data Rate}}
|2001, 2003
| rowspan="6" |[[Non-return-to-zero|NRZ]]
| rowspan="3" |[[8b/10b encoding|8b/10b]]<ref>{{cite web |title=InfiniBand Types and Speeds |url=https://www.advancedclustering.com/act_kb/infiniband-types-speeds/}}</ref>
|2.5
|2
|'''8'''
|16
|24
|5
|-
!{{abbr|DDR|Double data rate}}
|2005
|5
|4
|'''16'''
|32
|48
|2.5
|-
!{{abbr|QDR|Quad Data Rate}}
|2007
|10
|8
|'''32'''
|64
|96
|1.3
|-
!{{abbr|FDR10|Fourteen Data Rate, 10 Gbit/s per lane}}
|2011
| rowspan="4" |[[64b/66b encoding|64b/66b]]
|10.3125<ref>{{Cite web |title=Interfaces |url=https://docs.nvidia.com/networking/display/SB77X0EDR/Interfaces |access-date=2023-11-12 |website=NVIDIA Docs |language=en |quote=FDR10 is a non-standard InfiniBand data rate, where each lane of a 4X port runs a bit rate of 10.3125&nbsp;Gbit/s with a 64b/66b encoding, resulting in an effective bandwidth of 40&nbsp;Gbit/s. FDR10 supports 20% more bandwidth over QDR due to better encoding rate.}}</ref>
|10
|'''40'''
|80
|120
|0.7
|-
!{{abbr|FDR|Fourteen Data Rate}}
|2011
|14.0625<ref>{{Cite web |date=2018-04-29 |title=324-Port InfiniBand FDR SwitchX® Switch Platform Hardware User Manual |url=https://network.nvidia.com/pdf/user_manuals/SX6518_User_Manual.pdf |access-date=2023-11-12 |website=nVidia |at=section 1.2 |quote=InfiniBand FDR and FDR10 Overview [...] FDR, standard InfiniBand data rate, where each lane of a 4X port runs a bit rate of 14.0625&nbsp;Gbit/s with a 64b/66b encoding, resulting in an effective bandwidth of 54.54&nbsp;Gbit/s. The FDR physical layer is an IBTA specified physical layer using different block types, deskew mechanism and framing rules. The SX6518 switch also supports FDR10, a non-standard InfiniBand data rate, where each lane of a 4X port runs a bit rate of 10.3125&nbsp;Gbit/s with a 64b/66b encoding, resulting in an effective bandwidth of 40&nbsp;Gbit/s.}}</ref><ref name="fdr_fact_sheet" />
|13.64
|'''54.54'''
|109.08
|163.64
|0.7
|-
!{{abbr|EDR|Enhanced Data Rate}}
|2014<ref name="ib_roadmap">{{Cite web |title=InfiniBand Roadmap - Advancing InfiniBand |url=https://www.infinibandta.org/infiniband-roadmap/ |website=InfiniBand Trade Association |language=en-US}}</ref>
|25.78125
|25
|'''100'''
|200
|300
|0.5
|-
!{{abbr|HDR|High Data Rate}}
|2018<ref name="ib_roadmap" />
| rowspan="3" |[[Pulse-amplitude modulation|PAM4]]
|53.125<ref>{{Cite web |title=Introduction |url=https://docs.nvidia.com/networking/display/ConnectX6VPI/Introduction |access-date=2023-11-12 |website=NVIDIA Docs |language=en}}</ref>
|50
|'''200'''
|400
|600
|<0.6<ref>https://www.mellanox.com/files/doc-2020/pb-connectx-6-vpi-card.pdf {{Bare URL PDF|date=March 2022}}</ref>
|-
!{{abbr|NDR|Next Data Rate}}
|2022<ref name="ib_roadmap" />
|256b/257b{{efn-lr|Using Reed-Solomon [[forward error correction]]}}
|106.25<ref>{{Cite web |title=Introduction |url=https://docs.nvidia.com/networking/display/ConnectX7VPI/Introduction |access-date=2023-11-12 |website=NVIDIA Docs |language=en}}</ref>
|100
|'''400'''
|800
|1200
| {{dunno}}
|-
!{{abbr|XDR|Extended Data Rate}}
|2024<ref>{{Cite web |title=NVIDIA Announces New Switches Optimized for Trillion-Parameter GPU Computing and AI Infrastructure |url=http://nvidianews.nvidia.com/news/networking-switches-gpu-computing-ai |access-date=2024-03-19 |website=NVIDIA Newsroom |language=en-us}}</ref>
| rowspan="2" | {{TBD}}
|200
|200
|'''800'''
|1600
|2400
| rowspan="2" | {{TBD}}
|-
!{{abbr|GDR|t.b.d. Data Rate}}
|{{TBA}}
|{{TBD}}
|400
|400
|'''1600'''
|3200
|4800
|}
; Notes
{{Notelist-lr}}

Each link is duplex. Links can be aggregated: most systems use a 4 link/lane connector (QSFP). HDR often makes use of 2x links (aka HDR100, 100&nbsp;Gb link using 2 lanes of HDR, while still using a QSFP connector). 8x is called for with NDR switch ports using OSFP (Octal Small Form Factor Pluggable) connectors {{Cite web |url=https://docs.nvidia.com/networking/display/cablemanagfaq/cable+and+connector+definitions|title=Cable and Connector Definitions | language=en-US}}

InfiniBand provides [[remote direct memory access]] (RDMA) capabilities for low CPU overhead.

=== Topology ===

InfiniBand uses a [[switched fabric]] topology, as opposed to early shared medium [[Ethernet]]. All transmissions begin or end at a channel adapter. Each processor contains a host channel adapter (HCA) and each peripheral has a target channel adapter (TCA). These adapters can also exchange information for security or [[quality of service]] (QoS).

=== Messages ===

InfiniBand transmits data in packets of up to 4&nbsp;KB that are taken together to form a message. A message can be:
* a remote direct memory access read or write
* a [[Communication channel|channel]] send or receive
* a transaction-based operation (that can be reversed)
* a [[multicast]] transmission
* an [[atomic operation]]

=== Physical interconnection ===
[[Image:Infinibandport.jpg|thumb|x100px|right|InfiniBand switch with CX4/SFF-8470 connectors]]

In addition to a board form factor connection, it can use both active and passive copper (up to 10 meters) and [[optical fiber cable]] (up to 10&nbsp;km).<ref name=faq>{{cite web|title=Specification FAQ|url=http://www.infinibandta.org/content/pages.php?pg=technology_faq|publisher=ITA|access-date=30 July 2014|archive-url=https://web.archive.org/web/20161124000007/http://infinibandta.org/content/pages.php?pg=technology_faq|archive-date=24 November 2016|url-status=dead}}</ref> 
[[QSFP]] connectors are used.

The InfiniBand Association also specified the [[CXP (connector)|CXP]] connector system for speeds up to 120&nbsp;Gbit/s over copper, active optical cables, and optical transceivers using parallel multi-mode fiber cables with 24-fiber MPO connectors.{{citation needed|date=August 2017}}

=== Software interfaces ===
Mellanox operating system support is available for [[Solaris (operating system)|Solaris]], [[FreeBSD]],<ref>{{cite web|title=Mellanox OFED for FreeBSD|url=http://www.mellanox.com/page/products_dyn?product_family=193|publisher=Mellanox|access-date=19 September 2018}}</ref><ref>{{cite web |author1=Mellanox Technologies |title=FreeBSD Kernel Interfaces Manual, mlx5en |url=https://www.freebsd.org/cgi/man.cgi?query=mlx5en |website=FreeBSD Man Pages |publisher=FreeBSD |access-date=19 September 2018 |language=en |date=3 December 2015}}</ref> [[Red Hat Enterprise Linux]], [[SUSE Linux Enterprise Server]] (SLES), [[Windows (operating system)|Windows]], [[HP-UX]], [[VMware ESX]],<ref>{{cite web|title=InfiniBand Cards - Overview|url= http://www.mellanox.com/page/infiniband_cards_overview|publisher= Mellanox|access-date= 30 July 2014}}</ref> and [[AIX]].<ref>{{cite web|title=Implementing InfiniBand on IBM System p (IBM Redbook SG24-7351-00)|url=http://www.redbooks.ibm.com/redbooks/pdfs/sg247351.pdf}}</ref>

InfiniBand has no specific standard [[application programming interface]] (API). The standard only lists a set of verbs such as <code>ibv_open_device</code> or <code>ibv_post_send</code>, which are abstract representations of functions or methods that must exist. The syntax of these functions is left to the vendors. Sometimes for reference this is called the ''verbs'' API. The [[de facto standard]] software is developed by [[OpenFabrics Alliance]] and called the Open Fabrics Enterprise Distribution (OFED). It is released under two licenses [[GPL2]] or [[BSD license]] for Linux and FreeBSD, and as Mellanox OFED for Windows (product names: WinOF / WinOF-2; attributed as host controller driver for matching specific ConnectX 3 to 5 devices)<ref>[http://www.mellanox.com/page/products_dyn?product_family=32&menu_section=34 Mellanox OFED for Windows - WinOF / WinOF-2]</ref> under a choice of BSD license for Windows. 
It has been adopted by most of the InfiniBand vendors, for [[Linux]], [[FreeBSD]], and [[Microsoft Windows]]. [[IBM]] refers to a software library called <code>libibverbs</code>, for its [[AIX]] operating system, as well as "AIX InfiniBand verbs".<ref>{{Cite web |title= Verbs API |work= IBM AIX 7.1 documentation |url= https://www.ibm.com/support/knowledgecenter/en/ssw_aix_71/com.ibm.aix.rdma/verbs_API.htm |date= 2020 |access-date= September 26, 2021 }}</ref>
The Linux kernel support was integrated in 2005 into the kernel version 2.6.11.<ref>{{Cite web |title= Verbs programming tutorial |date= March 11, 2014 |author= Dotan Barak |publisher= Mellanox |work= OpenSHEM, 2014 |url= https://www.csm.ornl.gov/workshops/openshmem2014/documents/presentations_and_tutorials/Tutorials/Verbs%20programming%20tutorial-final.pdf |access-date= September 26, 2021 }}</ref>

=== Ethernet over InfiniBand ===

Ethernet over InfiniBand, abbreviated to EoIB, is an Ethernet implementation over the InfiniBand protocol and connector technology.
EoIB enables multiple Ethernet [[Bandwidth (computing)|bandwidths]] varying on the InfiniBand (IB) version.<ref>{{cite web|title=10 Advantages of InfiniBand | url=https://www.naddod.com/blog/top-10-advantages-of-infiniband |website=NADDOD|access-date=January 28, 2023}}</ref>
Ethernet's implementation of the [[Internet Protocol Suite]], usually referred to as TCP/IP, is different in some details compared to the direct InfiniBand protocol in IP over IB (IPoIB).

{| class="wikitable sortable" 
|-
|+ Ethernet over InfiniBand performance
! Type !! Lanes !! Bandwidth (Gbit/s) !! Compatible Ethernet type(s) !! Compatible Ethernet quantity
|-
! rowspan="4" | SDR 
| {{0|00}}1 || {{0|000}}2.5 || GbE to 2.5 GbE || {{0}}2 × GbE to 1 × {{0}}2.5 GbE
|-
| {{0|00}}4 || {{0|00}}10 || GbE to 10 GbE || 10 × GbE to 1 × 10 GbE
|-
| {{0|00}}8 || {{0|00}}20 || GbE to 10 GbE || 20 × GbE to 2 × 10 GbE
|-
| {{0}}12 || {{0|00}}30 || GbE to 25 GbE || 30 × GbE to 1 × 25 GbE + 1 × {{0}}5 GbE
|-
! rowspan="4" | DDR 
| {{0|00}}1 || {{0|000}}5 || GbE to 5 GbE || {{0}}5 × GbE to 1 × {{0}}5 GbE
|-
| {{0|00}}4 || {{0|00}}20 || GbE to 10 GbE || 20 × GbE to 2 × 10 GbE
|-
| {{0|00}}8 || {{0|00}}40 || GbE to 40 GbE || 40 × GbE to 1 × 40 GbE
|-
| {{0}}12 || {{0|00}}60 || GbE to 50 GbE || 60 × GbE to 1 × 50 GbE + 1 × 10 GbE
|-
! rowspan="2" | QDR 
| {{0|00}}1 || {{0|00}}10 || GbE to 10 GbE || 10 × GbE to 1 × 10 GbE
|-
| {{0|00}}4 || {{0|00}}40 || GbE to 40 GbE || 40 × GbE to 1 × 40 GbE
|}