Micron Unveils Ultra-Dense LPDRAM

Micron just unveiled the world's first high-capacity 256GB LPDRAM SOCAMM2 module, a huge leap for memory density in data centers. The new memory triples performance-per-watt and cuts power consumption by two-thirds compared to previous solutions, directly addressing major bottlenecks in AI hardware and embedded systems.

This module is built on the industry's first monolithic 32Gb LPDDR5X die, enabling its massive 256GB capacity. This high density allows for up to 2TB of LPDRAM per 8-channel server CPU, a critical factor for handling the large model parameters and expansive context windows required by advanced AI inference workloads. The move to the CAMM2 (Compression Attached Memory Module) form factor is a significant shift from traditional SODIMMs and soldered-down memory. Standardized by JEDEC in late 2023, CAMM2 uses a land grid array (LGA) connector instead of an edge connector, which shortens the signal path to the CPU. This design improves signal integrity, enabling the higher data rates (up to 9600 MT/s) and lower latency necessary for AI processing. Power efficiency is a major driver for adopting LPDRAM in data centers, which globally consumed around 460 TWh in 2022. LPDDR5X memory can consume up to 77% less power compared to DDR5. This is achieved through lower operating voltages (around 1.05V for LPDDR5 vs 1.1V for DDR5) and advanced power management modes. For power-constrained data centers grappling with the high energy demands of AI servers, this reduction is crucial. The SOCAMM2 standard, an evolution of Dell's original proprietary CAMM, makes low-power DRAM modular and upgradeable for the first time in this class of device. This serviceability is a key advantage for enterprise IT and data center operators, who previously had to choose between the performance of soldered LPDDR and the modularity of slower SODIMMs. Micron's sampling of these modules to customers aligns with the expected ramp-up of next-generation AI server products, including a collaboration with NVIDIA. The adoption of the JEDEC CAMM2 standard ensures that other manufacturers can also produce compatible modules, fostering a broader ecosystem beyond a single vendor.

Get your own daily briefing

Scout delivers personalized news, insights, and conversations tailored to your role and industry.

Download on the App Store

Shared from Scout - Be the smartest in the room.