Micron delivers its highest-capacity SOCAMM2 to customers

Micron has sampled its first 192GB LPDDR5X SOCAMM2 memory modules to customers

Micron has delivered its first 192GB SOCAMM2 memory samples to customers, delivering users a 50% memory capacity increase over its last generation modules. With LPDDR5X memory, using Micron’s advanced 1-gamma DRAM process, Micron also promises 20% improvements to power efficiency over last-generation SOCAMM modules. This enables increased performance and support for larger data sets. Both of these changes are vital for future AI workloads.

For full-rack AI installation, Micron says that these new memory modules can provide customers with over 40 terabytes of CPU-attached low-power DRAM. Thanks to the modular design of SOCAMM2 modules, users have highly serviceable memory modules and the ability to deliver future capacity upgrades. If/when higher capacity SOCAMM2 modules launch, users can upgrade their servers with ease.

SOCAMM2 was built thanks to Micron’s long-term partnership with Nvidia, and is on its way to becoming an official JEDEC memory specification. Note that this shift will allow other DRAM manufacturers to launch SOCAMM2 memory modules. However, Micron will be first to market.

Thanks to SOCAMM2, LPDDR5X memory (which is typically used in mobile devices) is coming to datacenters. Now, Micron is leveraging this memory to deliver higher levels of performance and power efficiency in AI datacenters.

Micron is now shipping SOCAMM2 modules to customers with up to 192 GB capacities. Module speeds of up to 9.6 Gbps are shipping. Note that Micron plans to achieve higher volume SOCAMM2 production soon. This ramp-up will align with “customer launch schedules”. Note that this ramp will likely align with Nvidia’s ramp for its next-generation AI server products.

You can join the discussion on Micron’s 192GB SOCAMM2 memory modules on the OC3D Forums.