Samsung DDR5 CXL 2.0 256GB DIMM (CMM-D)

The Samsung DDR5 CXL 2.0 256GB DIMM (CMM-D), manufactured part number MCMQ16G82AMFA-DYAAT, is a Compute Express Link 2.0 memory expansion module built on Samsung's advanced DDR5 DRAM architecture. The CMM-D (CXL Memory Module – DRAM) form factor connects to a host CPU via a PCIe 5.0 x8 interface using the CXL 2.0 protocol, enabling memory capacity expansion beyond the physical DIMM slot limitations of standard server platforms. This approach allows datacenter architects to decouple memory capacity from CPU socket count, addressing the memory wall challenge that constrains large-scale in-memory computing environments.

Operating as a CXL Type 3 memory device, the CMM-D exposes its full 256GB of DDR5 DRAM as host-managed device memory (HDM), making it directly accessible to the host processor's memory subsystem without requiring a separate compute element on the device. The module leverages DDR5's inherent advantages — including improved power efficiency per bit compared to DDR4 and higher bandwidth per channel — while the CXL 2.0 interface provides a coherent, low-latency memory expansion path that is significantly closer to native DRAM latency than traditional storage-class or networked memory solutions. The module operates in a standard EDSFF or PCIe add-in card form factor designed for enterprise rack servers.

The Samsung CMM-D is engineered for the most demanding enterprise and hyperscale workloads where per-socket DRAM capacity is the critical bottleneck. AI and machine learning inference serving, large language model (LLM) deployment, in-memory databases, real-time analytics, and high-performance computing applications all benefit directly from the ability to present a single server node with substantially more addressable memory than physical DIMM slots allow. Samsung's enterprise-grade qualification and reliability standards make the CMM-D suitable for continuous 24/7 datacenter operation across environments spanning cloud infrastructure, AI accelerator nodes, and enterprise database servers throughout the UAE, GCC, EMEA, and APAC regions.

Ideal for

• Large Language Model (LLM) inference serving: expanding per-node memory capacity to hold multi-hundred-billion-parameter model weights entirely in DRAM, eliminating weight-offloading latency on AI inference servers
• In-memory database scaling: enabling platforms such as SAP HANA, Redis, and VoltDB to maintain substantially larger hot datasets in addressable memory on a single server node, reducing query latency and inter-node data movement
• Real-time analytics and business intelligence: supporting high-throughput analytical query engines that require large working-set memory to perform in-memory aggregations and joins across enterprise-scale datasets without spilling to NVMe
• High-Performance Computing (HPC) and scientific simulation: providing memory-bound simulation workloads — including computational fluid dynamics, genomics pipelines, and financial risk modeling — with expanded addressable capacity to increase problem-size per node
• Virtualization and cloud-native infrastructure: allowing hypervisor hosts to run higher VM densities or provision memory-intensive virtual machines and containers without adding additional physical servers, improving rack-level utilization
• Datacenter memory tiering and disaggregation: serving as a CXL-attached memory tier in composable infrastructure architectures where memory resources are dynamically allocated across workloads, supporting next-generation software-defined datacenter designs

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.