Intel announces the Compute Express Link (CXL)-- an interconnect technology designed to improve performance and remove bottlenecks between CPUs and other system components, such as GPUs, memory and FPGAs.
The result of 4 years of development by Intel, CXL is an open technology belonging to the CXL Consortium, an industry body formed in collaboration Alibaba, Cisco, Dell EMC, Facebook, Google, Hewlett Packard Enterprise, Huawei and Microsoft with the aim to promote and push adoption of the technology.
But what about the actual interconnect? According to Intel, CXL creates a high-speed, low latency connection between the CPU and workload accelerators such as GPUs, FPGAs and networking. It promises to maintain memory coherency between devices, allowing resource sharing for higher performance, reduced software stack complexity and lower system costs. Importantly, it runs across the PCIe 5.0 bus, making it electrically and mechanically compatible with the upcoming interface.
Intel adds the first generation CXL spec will be available to consortium members in H1 2019, before the release of products incorporating the technology in 2021. The first of such products will be aimed at enterprise, and will include Xeon processors, FPGAs, GPUs and Smart NICs.
