Micron Announces 128 GB DDR5 Module For Memory-Intensive Apps Like Generative AI

Semiconductor manufacturer Micron Technology has announced its 32GB monolithic die-based 128 GB DDR5 RDIMM memory module with performance of up to 8,000 MT/s (megatransfers per second) to support data centre workloads.

These high-capacity, high-speed memory modules are designed to meet performance and data handling requirements of a range of applications in data centre and cloud environments, Micron said. Applications include artificial intelligence, in-memory databases and processing for multithreaded, multicore count general compute workloads.

According to Micron, the modules are powered by its 1β technology and—compared to 3DS through-silicon via products—deliver around 45% improved bit density, up to 24% improved energy efficiency, 16% lower latency, and 28% improvement in AI training performance.

The 128GB DDR5 RDIMMs can deliver the memory bandwidth and capacity required for increasingly compute-intensive workloads, said Praveen Vaidyanathan, vice president and general manager of Micron’s Compute Products Group, adding that it will enable improvements to the data centre ecosystem.

The module uses die architecture choices for leading array efficiency and the densest monolithic DRAM die, Micron said. Voltage domain and refresh management features help optimise the power delivery network and improve energy efficiency.

By leveraging AI-powered manufacturing methods, the 1β process technology node has achieved the fastest yield maturity, as compared to 1α line, Micron said. The 128GB RDIMMs will be shipping in platforms capable of 4,800 MT/s, 5,600 MT/s and 6,400 MT/s and will be compatible with future platforms capable of up to 8,000MT/s.

According to Micron, the 32GB DRAM die enables future expansion of the memory portfolio with enhanced bandwidth and energy-efficient MCRDIMM and JEDEC standard MRDIMM products in 128GB, 256GB, and higher capacity solutions. This will enable enterprises to integrate optimised solutions for AI and high-performance computing applications that suit their needs for bandwidth, capacity, and power optimisation, the company said.