Samsung is laying claim to the first-ever 512GB
Compute Express Link (CXL) memory module, which packs four times the DRAM as its previous version. In doing so, Samsung provides an upgrade path to scale servers to massive amounts of memory—tens of terabytes of DRAM—with just one-fifth of the system latency.
That's what the
CXL memory module form factor is all about. It allows clients to scale memory capacities and bandwidth in data centers to terabyte levels by leveraging an open, industry-supported interconnected based on the high-speed PCI Express 5.0 bus. These modules are based on DDR5 RAM and get recognized as main memory by both the CPU and GPU.
In addition to essentially forming an end-around to the capacity limits imposed by memory channels, Samsung's CXL-enabled modules also dramatically reduce system latency caused by memory caching, and increase bandwidth to several terabytes per second..
"CXL DRAM will become a critical turning point for future computing structures by substantially advancing artificial intelligence (AI) and big data services, as we aggressively expand its usage in next-generation memory architectures including software-defined memory (SDM)," said Cheolmin Park, Vice President of Memory Global Sales & Marketing at Samsung Electronics, and Director of the CXL Consortium.
Samsung sees big market opportunity for its CXL modules with the emerging metaverse and related experiences, AI driven workloads, and big data "generating explosive amounts of data."
The upgraded 512GB CXL module will be the first memory device to support PCIe 5 and come in an EDSFF (E3.2) form factor. In addition, Samsung says it is getting ready to unveil an updated version of its open source Scalable Memory Development Kit (SMDK). Samsung describes it as a comprehensive toolkit that allows the CXL memory expander to work seamlessly in heterogeneous memory systems.
Samsung will being
sampling 512GB CXL DRAM to customers in the third quarter of this year, and expects to have it ready for commercial adoption when next-gen platforms arrive.
