Intel QuickPath Interconnect

from Wikipedia, the free encyclopedia

QuickPath Interconnect ( QPI for short) is a point-to-point connection developedby Intel for communication between processors and for communication between processors and chipset . In contrast to its predecessor Front Side Bus (FSB),QPI isnot a bus system , but a routing mechanism that ensures that data packets in a network made up of several processors reach the correct recipient CPU. QPI was developed as a Common System Interface ( CSI ) and replacedthe FSB in Intel-based systems fromthe Core i7 and the associated introduction of the Nehalem architecture .

Technical

Clock
frequency
Transfers net Gross aggregated
gross
year
(GHz) (GT / s) Data rate (GByte / s)
QPI
002.4 004.8 009.6 012 0024 2008
002.93 005.86 011.73 014.66 0029.33 2009
003.2 006.4 012.8 016 0032 2008
003.6 007.2 014.4 018th 0036 2012
004th 008th 016 020th 0040 2012
004.8 009.6 019.2 024 0048 2014
UPI
004.8 009.6 019.2 024 0048 2017
005.2 010.4 020.8 026th 0052 2017

QPI was developed in competition with AMD's HyperTransport technology and the first version has been in production since the second half of 2008.

A QPI port consists of two unidirectional QPI links, one for each transmission direction . A QPI link consists of 20 wire pairs for the symmetrical transmission of 20 bits. QPI of the first version was clocked at 2.4 or 3.2 GHz and operated in the DDR method with two transfers per clock or one transfer per clock edge, which translates into a transfer rate of 4.8 or 6.4  gigatransfer per second ( GT / s).

Since QPI is a full duplex point-to-point connection, two (unidirectional) links are required for a QPI connection.

However, considering the bandwidth alone is not indicative of the speed of a system. In addition, there is latency as an equally important criterion. With the FSB it is probably lower because it is a bus system and there is no routing here. The crossbar router in the new Nehalem processors is an additional component on the transmission path that increases the signal propagation times and thus the latency, which does not affect the data rate, but lowers the speed of the overall system. Which is faster in practice, FSB1600 or QPI of the first version, depends on the application. If small amounts of data are often transmitted one after the other, FSB has an advantage due to the lower latency, and QPI for large amounts of data in one go. Since QPI can transmit bidirectionally, the disadvantage of the higher latency is compensated by the FSB, provided that the application requires it, which is certainly the case with multi-core processors, if z. B. one core fetches data from the memory and the other transmits data to the graphics card at the same time.

The transmission on a QPI link takes place in 80-bit packets (called Flit by Intel), for which a 4-bit transfer or transmission cycle (called phits) is required for 20 line pairs. Of the 80 bits of a flit, 64 bits are user data, the remaining 16 bits are meta data. Metadata contain CRC error correction information as well as routing data for the link layer for message classes and virtual channels .

Until March 31, 2010, a QPI clock of 3.2 GHz was reserved for the larger Core i7 models only. On this day, however, Intel announced that the Core i7 920, and the other i7 CPUs with a QPI of 4.8 GT / s, can also be operated at 6.4 GT / s without loss of warranty. This applies to all models, including the older C0 stepping.

See also

Web links

Individual evidence

  1. hardware-infos.com: all Core i7s with 6.4 GT / s without loss of warranty