Micron and Intel are partnering to bring to market multiplexed dual in-line memory modules (MRDIMMs), an innovative technology designed to deliver higher memory bandwidth and lower latency for data center and high-performance computing (HPC) applications. Micron recently announced that it has sampled MRDIMMs, which offer up to 256GB of capacity and 40% lower latency than traditional RDIMMs, providing powerful support for AI inference and high-performance computing applications. MRDIMM modules support a wide range of capacity options from 32GB to 256GB, which makes them ideal for memory-intensive applications such as AI inference and HPC workloads, which typically require more than 128GB of memory per DIMM slot.
MRDIMM technology significantly increases the bandwidth of memory by combining two DDR5 DIMMs to deliver double the data rate to the host. This design allows access to two columns at the same time (Rank), allowing the memory module to transfer data at a higher rate. For example, if two DDR5 DIMMs are combined at 4400MT/s, the output will be 8800MT/s. The MRDIMM module provides seamless compatibility with existing processor platforms through DDR5 interface and technology, while providing higher bandwidth, lower latency, and a variety of capacity specifications. Compared with RDIMM, MRDIMM has up to 39% more effective memory bandwidth and more than 15% bus efficiency.
Pictured: Micron and Intel unveil MRDIMM modules
Compatible with Intel's Xeon 6 processors, Micron's MRDIMM offers a wide range of capacities from 32GB to 256GB and covers both standard and high-form factor (TFF) for high-performance 1U and 2U servers. This high-profile form factor (TFF) module optimizes thermal design to reduce DRAM temperatures by up to 20 degrees Celsius with the same power consumption and airflow, resulting in more efficient cooling for data centers and optimizing total system task energy consumption for memory-intensive workloads. Suitable for a wide range of workloads.
Matt Langman, vice president at Intel, said that MRDIMM enables seamless compatibility with existing Intel Xeon 6 processor platforms, giving customers more flexibility and choice. Scott Tease, vice president of Lenovo Group, also pointed out that with the exponential growth of processor cores, MRDIMM will help close the memory bandwidth gap and meet the needs of system performance balance.
It is reported that Micron MRDIMM is now on sale and is expected to be shipped in batches in the second half of 2024. Subsequent generations of MRDIMM products will continue to deliver 45% higher single-channel memory bandwidth than their RDIMM generation. This indicates that MRDIMM technology is expected to be the preferred solution for the main memory of AI servers and supercomputer systems in the future, especially in scenarios where there are specific requirements for large capacity, high-cost performance, and scalability.
