CMM-D 128GB PCIe E3.S CXL2.0

General Description

CXL Memory Module - DRAM (CMM-D) is a Compute Express Link™ (CXL™) based Memory Module Device developed to meet growing demands for memory bandwidth and capacity in applications such as artificial intelligence, machine learning, cloud infrastructure and high performance computing. 

CXL provides efficient connectivity between the host CPU and connected devices such as accelerators and memory expansion devices.

CMM-D is a CXL Type3 device supporting CXL.mem (responding to the memory read/write requests from CPU host) and CXL.io (device discovery, enumeration, error reporting, and management) protocols which have been optimized for low latency.

Memory for AI Era

CXL is next-generation interface with enhanced scalability and bandwidth. CMM-D is optimal solution for effectively processing the exploding amount of data that has resulted from advancements in AI/ML.

Traditionally, adding memory capacity and bandwidth in a system involves increasing the number of native CPU memory channels. But adding memory channels to a CPU increases engineering complexity and drives up cost. 

A CXL Type 3 memory expansion device provides a flexible and powerful option to increase memory capacity and increase memory bandwidth, without increasing the number of primary CPU memory channels.

By combining the high-performance Samsung DRAM DDR5 16Gb 5200 silicon with DRAM management technologies, the Samsung CXL Memory Module delivers the extended performance of up to 32GT/s, which is suitable for enterprise applications.

More Scalability

The existing DRAM DDR interface has limitations when it comes to expansion, due to the restrictions of DRAM-module capacity and the number of DDR slots. To deal with this, there have traditionally been two main methods to further expand memory. 

First, it is possible to configure multiple memory nodes that consist of both CPU and memory. Second, if the host runs out of memory, Remote Direct Memory Access (RDMA) can be used to utilize memory from network-connected memory nodes.

However, these approaches have drawbacks. Configuring additional memory nodes increases costs by adding more CPUs and systems. RDMA, on the other hand, can decrease performance due to the software overhead of data transfer. 

CMM-D solves both these problems by reducing overhead and enabling DRAM to be mounted to the PCIe interface, where SSDs are usually connected. This allows it to significantly expand memory capacity without the introduction of additional systems.

Smart Memory Tiering - Memory Promotion, Demotion

Non-Uniform Memory Access (NUMA) is a method for efficiently managing memory. It resolves bottlenecking by separating the areas of memory that are accessible by each CPU.

NUMA supported operating system can now take full advantage of CXL memory, maximizing the utilization of CMM-D – even when it is not dedicated to a specific CPU.

When seldom-accessed data is stored in local memory, it reduces the performance of frequently used (or “hot”) memory. Fortunately, this problem can now be effectively solved by using CXL memory with RHEL's memory tiering function. 

This function allocates hot memory to the local memory tier and infrequently-used memory to the CXL tier, allowing for data migration between tiers when necessary.

More features

CMM-D includes key CXL™ 2.0 Reliability, Availability and Serviceability (RAS) features, such as error logging which allows the host to perform RAS operations, like page retirement without dependence on a driver.