PCIe 3.0 x16 and QPI 1.1 (20 lanes) have the same effective bandwidth (16 GB / s). So, I wanted to get a rough picture of the differences between them.
What is the difference between these two factors in terms of latency and message rate (number of packets or TLP per second)? For latency, my ball numbers are 20 ns for QPI and 200 ns for PCIe 3.0. Are these good grades? If so, why is the PCIe delay much higher - is it because of the length of the wire?
Are there significant architectural differences between the two, besides QPI that provides cache tracking? As far as I know, both use a layered protocol: the transport layer through the physical layer.
- . QPI MESIF NUMA . PCIe , (. PCIe). , . , - , QPI PCIe ( , QPI Intel).
GT/s. QPI: 4.8, 6.4 8 GT/s 5 8 GT/s PCIe.
, , . QPI . , 4+ QPI . PCIe 500ns, . PCIe, ( PEG), , , .
Source: https://habr.com/ru/post/1544402/More articles:The FORM-CONTROL class in Bootstrap does not take effect when used in datatables - javascriptHow to configure Jenkins Gerrit Trigger to exclude file path? - jenkinsHow can I make sure that I am implementing the Open / Closed principle? - c #Can I request current voltages from Windows 7? - windows-7Why is the project format from Xcode 5.1.1 from Plist to XML? - gitXcode not autocomplete Swift analysis functions - autocompleteHow to get user input before saving a file in Sublime Text - sublimetextWhere is IsInstanceOfType or its equivalent in the new .NET Type / TypeInfo API? - c #Git pull with unstated changes - gitWhy the rails do not reset the test database between runs - ruby-on-railsAll Articles