Compute Express Link (CXL)

2019: Announcement

On March 11, 2019 Intel provided new technology of component interconnections Compute Express Link which will allow servers to use more effectively the video cards programmed by the user gate arrays (FPGA) and other accelerators optimized for special workloads.

Intel positions Compute Express Link as the new industry standard which will be supported by the equipment from different producers. For promotion of technology its development was transferred to new consortium which part the set of the large technology companies, including Cisco, Google and, of course, Intel was.

Intel submitted the standard for contact of server processors with accelerators

Compute Express Link is intended for connection of the central processor with the additional chips accelerators installed in the server. The companies quickly implement any accelerators, especially graphic, for acceleration of specialized processes, such as processing of algorithms of artificial intelligence.

Compute Express Link is based on the PCI Express 5.0 standard for connection of server components. Such bus allows to transfer data at a speed of 128 Gbytes / with in certain configurations.

Intel made a number of changes to use this speed for communication between chips in data centers. The mechanism which allows the processor to share memory with accelerators became one of key features of Compute Express Link. Such technology, according to developers, replaces the separate equipment which usually is required for accomplishment of this task, and, thus, increases system effectiveness.^[1]

The Compute Express Link interface will be open and it is not necessary to pay regular assignments for its use. Specifications of technology will be available to participants of consortium later in the first half  of 2019. In the 2021st support of the standard should appear in server platforms of Intel.

Notes