A consortium led by Cisco, Intel, Dell, Oracle, Lenovo, HPE, IBM, Microsoft, Samsung, NVIDIA, and Google developed the Compute Express Link (CXL) protocol.
CXL is poised to transform the way servers use memory, resulting in significant cost savings and increased application performance. Memory accounts for roughly half of the cost of a typical data center server, and the biggest issue in maximizing application performance is the so-called “memory wall,” or the difficulty in allowing processors to access data at the same rate at which they can process it. The protocol and its potential impact on data centers are on the horizon, with implementation set to begin in 2024.
CXL protocol transfers data as memory load-store operations rather than IO transfers, allowing it to achieve faster data transfer speeds required for memory over the existing PCIe links in servers. CXL will expand the amount of memory that CPUs can address while also potentially reducing the amount of memory required in data centers.
According to IT industry analysts and directors, CXL will provide the necessary impetus to overcome memory’s high cost, difficulty in using it efficiently or at high utilization rates, and the limit on the amount of memory that an individual CPU can rapidly access.
Intel and AMD will deploy a new generation of CPUs that support CXL, and CXL evaluations are currently underway. The savings will undoubtedly be substantial, ensuring widespread adoption. CXL also addresses the stranded memory issue, in which data center operators must over-provision servers with memory capacity that is frequently unused. Stranded memory affects hyperscalers and other data center operators as a result of mixed and unpredictable workloads.
CXL also plans to introduce new rack-level data center architectures, which will significantly reduce data center costs while improving application performance. The protocol is expected to have a significant impact on data centers.
The sources for this piece include an article in DataCenterKnowledge.
