Micron’s Data Center MRDIMMs Double the Bandwidth of Existing DDR5 DIMMs – Promising Enhanced AI and HPC Performance with Xeon 6 Granite Rapid CPUs
Micron has confirmed that it is launching its new multiplexed rank dual inline memory module (MRDIMM). Designed for scenarios requiring more than 128GB of memory per DIMM slot, the new memory works by combining multiple layers of memory modules, running at standard DDR5 speeds, with a data buffer that accelerates the connection to the host computer, allowing for faster data transfers.
MRDIMMs improve upon traditional TSV RDIMMs by providing greater bandwidth, larger capacity, lower latency, and improved performance per watt.
Designed for use in virtualized environments, HPC and AI data center operations, the MRDIMMs are the first generation in Micron’s new memory module family and are compatible with Intel’s new Xeon 6 processors.
Used in Granite Rapids
“Micron’s latest innovative main memory solution, MRDIMM, delivers much-needed bandwidth and capacity with lower latency to scale AI inference and HPC applications on next-generation server platforms,” said Praveen Vaidyanathan, Vice President and General Manager of Micron’s Compute Products Group.
“MRDIMMs significantly reduce the amount of power used per task while offering the same reliability, availability, serviceability and interface as RDIMMs, giving customers a flexible solution that scales performance. Micron’s close industry partnerships enable seamless integration into existing server infrastructures and smooth transitions to future computing platforms.”
MRDIMM technology is based on the physical and electrical DDR5 standards and is intended to enable greater scalability of bandwidth and capacity per core. The modules offer up to 39% more effective memory bandwidth, over 15% improvement in bus efficiency, and up to 40% less latency, compared to RDIMMs.
Available in capacities from 32GB to 256GB and suitable for both standard and tall form factors, MRDIMMs are geared toward 1U and 2U high-performance servers. Micron says they feature an advanced thermal design that reduces DRAM temperatures by up to 20 degrees Celsius, helping to improve cooling efficiency and reduce overall power consumption for memory-intensive tasks.
Micron plans to begin mass shipping of MRDIMMs in the second half of 2024, timed to coincide with Intel’s launch of its 128 P-core Granite Rapids-AP line, a move that will see the chip giant overtake AMD as the fastest CPU server for the first time in a decade. Granite Rapids will be the first platform to support MRDIMMs with 8800MT/s on the host side.