[News] Next-Gen HBM? NVIDIA Reportedly Teams up with Samsung/ SK hynix to Push SOCAMM for 2025 Production

As NVIDIA collaborates closely with memory heavyweights on their latest HBM products for AI accelerators, the U.S. chip giant is said to be in talks with Samsung and SK hynix to push the commercialization of a next-gen memory technology called SOCAMM (System on Chip Advanced Memory Module), according to Korean media outlets EBN and Business Korea.

The EBN report indicates that NVIDIA and memory manufacturers are exchanging SOCAMM prototypes for testing, with mass production potentially starting by the end of 2025.

At CES 2025, NVIDIA unveiled its GB10 Grace Blackwell Superchip and Project DIGITS, aiming to popularize personal AI supercomputers. SOCAMM, seen as the “next-gen” HBM with superior performance and power efficiency over traditional DRAM for small PCs and laptops, could be the key, according to EBN.

Notably, the EBN report implies that NVIDIA plans to use individual LPDDR for the first product of its “DIGITS” line, with plans to incorporate four SOCAMM modules in the next version.

The reports highlight that unlike DDR4 and DDR5-based SODIMM modules, SOCAMM uses low-power LPDDR5X for better efficiency and performance. With significantly more I/O pins, it can dramatically boost data transfer speeds, which would be crucial for AI computing, the reports add.

In addition, the biggest strength of SOCAMM is its “detachable module” design, which allows users to easily replace the memory, according to the reports.

The reports also suggest NVIDIA’s push for its own memory standard marks a major shift from JEDEC’s traditional framework. While JEDEC includes memory giants like Samsung, SK hynix, and Micron, its members also span semiconductor, server, and PC firms like Arm, NXP, Intel, HP, and Honeywell.

Read more

• [News] SK hynix Reported to Deliver HBM4 Samples to NVIDIA in June, with Mass Production by Q3 2025
• [News] NVIDIA’s GB10 Superchip Powering Project DIGITS is Reportedly Built with TSMC’s 3nm Node

(Photo credit: NVIDIA)