Visit the purchase page:AMD Flagship Store
In March this year,Intel led the development of CXL technology and became the initial versionAfterwards, they joined Alibaba, Cisco,DellEMC, Facebook, Google,HPE,Huaweias well asMicrosoftEstablish CXL Alliance.
Previously,AMDDifferent organizations have been added to push to solve cache consistency problems in different systems and accelerators, including CCIX, OpenCAPI and Gen-Z.
It is reported that with the explosive growth of data and the rapid innovation of specific workloads, such as compression, encryption and artificial intelligence, the collaboration of dedicated accelerators and general CPUs in heterogeneous computing has been promoted. These accelerators need not only high-performance connections with processors, but ideally they can also share a common memory to reduce loss and latency.In the era of PCIe 5.0 high bandwidth, CXL will become a key technology to achieve more coherent memory sharing between accelerator and CPU.
Specifically, CXL creates high-speed and low-latency interconnections between CPU and workload accelerators (such as GPU, FPGA and network), enabling memory consistency among devices, allowing resource sharing, thus achieving higher performance, reducing software stack complexity, and lower overall system cost.
Although there are other interconnection protocols in the industry, CXL has demonstrated its unique value in providing CPU/device memory consistency, reducing device complexity, and providing industry-standard physical and electrical interfaces in a single technology to achieve the best plug-and-play experience.