Micron is the first to ship high-end DDR5 RDIMM memory for AI data centers, at up to 8000MT/s

Micron has released the industry's first high-end 128GB DDR5 RDIMM memory that's destined for AI data centers, with speeds of up to 8000MT/s.

2

Micron announced today that it has validated and shipped its high-capacity monolithic 32Gb DRAM die-based 128GB DDR5 RDIMM memory with speeds starting at 5600MT/s on all leading server platforms. The company explains that its new memory modules are powered by Micron's in-house, industry-leading 1-beta technology, with 45% improved bit density, up to 22% improved energy efficiency, and up to 16% lower latency over competitive through silicon via (TSV) products.

Micron's new high-capacity 128GB DDR5 RDIMMs deliver up to 28% faster performance for AI training, while the 16% less latency is important for memory-bound workloads including generative AI, in-memory databases, and real-time data analytics, where high capacity is needed, and prompt response times are critical for real-time inference.

Praveen Vaidyanathan, VP and GM of Micron's Compute Products Group, said: "With this latest volume shipment milestone, Micron continues to lead the market in providing high-capacity RDIMMs that have been qualified on all the major CPU platforms to our customers. AI servers will now be configured with Micron's 24GB 8-high HBM3E for GPU-attached memory and Micron's 128GB RDIMMs for CPU-attached memory to deliver the capacity, bandwidth and power-optimized infrastructure required for memory intensive workloads".

Dan McNamara, senior vice president and general manager, Server Business Unit, AMD, said: "A core tenet of our work with Micron is advancing the capabilities of data center infrastructure through highly-performant memory for compute intensive workloads. Through this collaboration, our joint customers can now get immediate impact out of the high-capacity DDR5 memory offering from Micron in an AMD EPYC CPU powered server, delivering the performance and efficiency needed for the modern data center".

Krista Satterthwaite, senior vice president and general manager, Compute at HPE, said: "Adopting advanced memory capabilities, while ensuring high-performance and efficiency, is critical to supporting growing AI workloads in training, tuning, and inferencing. We are committed to providing the most high-performing, energy-efficient solutions, and through our collaboration with Micron, plan to deliver monolithic, high-density DRAM across our AI portfolio to help our enterprise customers gain optimal performance to tackle any workload".

Dr. Dimitrios Ziakas, vice president of Intel's Memory and IO Technologies, Intel Corporation, said: "Micron's 128GB DDR5 RDIMM memory is the first 32Gb monolithic DRAM-based high-capacity DIMM that has completed Intel platform memory compatibility qualification on 4th and 5th Gen Intel® Xeon® processors. 32Gb density based DDR5 DIMMs accelerates critical server and AI system configurations bringing forward key performance, capacity, and most importantly power efficiency benefits to Intel® Xeon® processor-based systems. We are excited to continue our collaboration with Micron to drive broad adoption of innovative products in the market that solve memory capacity and power bottlenecks for AI and server customers".

Wally Liaw, senior vice president of Business Development and co-founder at Supermicro, said: "Supermicro is leading the industry with the broadest accelerated server and solution portfolio based on NVIDIA, AMD and Intel. Savvy customers are looking for large memory footprint, performance, and efficiency improvements in the AI infrastructure. Customers can benefit significantly from Supermicro's advanced GPU SuperServers with the new 32Gb monolithic DRAM-based 128GB memory, and we are excited to collaborate with Micron to enable this".