Over four years ago, Intel started to develop what is now known as Compute Express Link (CXL), an interface to coherently connect CPUs to all types of other compute resources. Over time, Intel collaborated with other industry behemoths, and early this year nine companies organized the CXL Consortium to jointly develop the technology as a new open standard. Over the past few months, dozens of additional companies have joined the consortium, and now the consortium itself has been formally incorporated this week, marking a major step in the development of CXL as an industry standard.

While incorporation itself doesn't change matters for CXL from a technical perspective, incorporating a group like the CXL Consortium is a fairly big deal, because this typically only happens with an industry standards group gets large enough and gains enough traction that its members are very confident the technology is soon to go into widespread use. This means that the CXL Consortium has been elevated to the same level as the USB-IF, VESA, and other standard groups. Which is to say, all signs point to CXL eventually winning the war of cache-coherent interconnects, and becoming a major, long-term industry standard.

Meanwhile, wasting no time, the newly-incorporated organization has named five additional members of its board of directors, and it has released version 1.1 of the CXL specification.

Support Growth & New BOD Members

Being a CPU-to-everything cache-coherent interconnect protocol, CXL competes in one way or another against such technologies as CCIX, Gen-Z, Infinity Fabric, NVLink, and OpenCAPI, so broad industry support is tremendously important for the technology. Originally founded by Alibaba, Cisco, Dell EMC, Facebook, Google, HPE, Huawei, Intel, and Microsoft, the CXL Consortium has gained over 50 additional members over the past few months. The consortium now counts nearly 70 companies and organizations in its ranks, from developers of CPUs, GPUs, FPGAs, SSDs, interconnects, servers, and other hardware as well as from software developers and cloud service providers.

Among the companies that recently joined the CXL Consortium are AMD, Arm, IBM, and Xilinx. To that end, the organization appointed five new members to its board of directors from AMD, Arm, IBM, Microchip, and Xilinx. The expanded board of directors now includes 13 members and looks as follows.

CXL Consortium: Members of the Board
Company Person Position
Alibaba Di Xu ?
AMD Nathan Kalyanasundharam Senior Fellow at AMD
Arm Dong Wei Standards Architect and Fellow
Cisco Sagar Borikar Principal Engineer, Data Center Systems Engineering
Dell EMC Kurtis Bowman Director of Technology and Architecture in Dell's Server CTO Office
Facebook Chris Petersen Hardware Systems Technologist
Google Rob Sprinkle Technical Lead for Platforms Infrastructure at Google
HPE Barry McAuliffe ?
IBM Steve Fields Fellow and Chief Engineer of Power Systems
Intel Jim Pappas Director of Technology Initiatives, Intel's Data Center Group.
Microchip Larrie Carr Fellow, Technical Strategy and Architecture, Data Center Solutions
Microsoft Leendert van Doorn  Distinguished Engineer
Xilinx Gaurav Singh Corporate Vice President

CXL 1.1 Published

Back in March, the nine founding members of the CXL Consortium published version 1.0 of the specification. By now, several refinements have been made, so this week the organization published version 1.1 of the spec. Unfortunately, the organization does not publicly disclose what changes it brings; though coming this soon after 1.0, it's likely little more than minor tweaks to address underdefined behavior and satisfy the needs of some of the new members.

As a refresher, CXL is designed to enable heterogeneous processing (by using accelerators) and memory systems (think memory expansion devices), the low-latency CXL runs on PCIe 5.0 PHY stack at 32 GT/s and supports x16, x8, and x4 link widths natively. CXL supports three protocols within itself: the mandatory CXL.io as well as CXL.cache for cache coherency, and CXL.memory for memory coherency that are needed to effectively manage latencies. When it comes to performance, a CXL-compliant device will enjoy 64 GB/s bandwidth in each direction when installed into a a PCIe 5.0 x16 slot. In addition, the protocol also supports degraded mode at 16.0 GT/s and 8.0 GT/s data rates as well as x2 and x1 links.

Related Reading:

Source: CXL Consortium

POST A COMMENT

5 Comments

View All Comments

  • Dug - Friday, September 20, 2019 - link

    Interesting that Intel just released Stratix 10 DX, which I guess is a stop gap until this comes out..
    I guess if you can't wait until 2021 when CXL is released, then this would be the solution, but seems short sighted.
    Reply
  • Smell This - Friday, September 20, 2019 - link

    Is this with the next, new omni-path? Was this tanked because of CXL? Reply
  • dersteffeneilers - Monday, September 23, 2019 - link

    No, Omni-Path was tanked bc it had no support from vendors other than Intel, and without that InfiniBand is clearly superior, even if you take the cost and energy savings from the fabric SKL-SP parts into account. Reply
  • DanNeely - Friday, September 20, 2019 - link

    I didn't realize Microchip made HPC products; I've always associated them with the PIC series of low power microcontrollers. Reply
  • TomWomack - Thursday, September 26, 2019 - link

    So will it finally be possible to build cache-coherent multiprocessors with chips from different manufacturers, where an Cortex-A79 can efficiently query the cache of an Icelake Xeon? Reply

Log in

Don't have an account? Sign up now