Home > Business > Tech

Samsung-Nvidia, SK hynix-Intel partnerships open new fronts in chip rivalry

Published: 19 Dec. 2025, 05:00

SK hynix’s 256GB DDR5 Rdimm, a high-capacity server dynamic random-access memory module based on 10-nanometer-class fifth-generation 32Gb chips [SK HYNIX]

Samsung Electronics and SK hynix are ramping up their rivalry in low-power server memory by partnering with U.S. chip titans Nvidia and Intel, respectively.

The two companies simultaneously unveiled their achievements: new high-capacity and high-performance commodity dynamic random-access memory (DRAM) products from SK hynix, and next-generation memory modules from Samsung.

With demand for memory surging thanks to AI, prices for high bandwidth memory (HBM) and conventional server DRAM have been driving up, leading to intensifying competition between Samsung and Sk hynix as the two grapple with the No. 1 position in the DRAM market.

SK hynix said on Thursday that its 256GB DDR5 registered dual in-line memory module (Rdimm) — based on 32Gb fifth-generation 10-nanometer-class (1b) DRAM — has passed Intel’s compatibility and performance verification process.

Rdimm is a memory module designed to ensure system stability even when multiple DRAM chips are installed. High-capacity products are mainly used in enterprise data centers.

“This product is the highest capacity Rdimm based on 1b 32Gb DRAM that is currently available, and it is the first Rdimm of this specification to pass Intel’s verification,” SK hynix said.

Although Intel has fallen behind Nvidia in GPUs, it still commands more than 70 percent of the data center CPU market. Now certified, SK hynix’s Rdimm can be immediately adopted in servers based on Intel’s latest Xeon 6 CPUs. Once Intel verifies the performance and compatibility of memory products, server makers such as Dell, HP and Supermicro typically purchase them in bulk for server production.

Samsung Electronics’ Socamm2, a low-power double data rate-based server memory module designed for AI data centers [SAMSUNG ELECTRONICS]

The new Rdimm from SK hynix is manufactured using through-silicon via technology, a core technology used in HBM. By stacking 32Gb DRAM chips in a 3-D structure and connecting them through microscopic vertical channels, the company achieved an industry-leading capacity of 256GB.

“Servers using this product deliver 16 percent higher AI inference performance while reducing power consumption by 18 percent compared to existing products,” SK hynix developers said.

Samsung Electronics, meanwhile, said on Thursday that it has completed development of a memory product optimized for AI data centers — small outline compression attached memory module (Socamm2) — and has supplied samples to Nvidia.

Socamm2 is a server memory module built by stacking low-power double data rate DRAM. Compared to the commonly used Rdimm, it is smaller in size, offers more than double the bandwidth and cuts power consumption by 55 percent, making it well suited for high-performance AI computing. Unlike HBM, which is packaged together with GPUs, Socamm2 is a removable module, making it easier to manufacture and offering greater design flexibility.

Socamm2 has yet to be commercialized, as international standardization and platform compatibility remain unresolved. Nvidia is reportedly planning to apply Socamm2 to its next-generation AI platform Vera Rubin, scheduled for release in the second half of next year.

A composite image of Samsung Electronics and SK hynix [YONHAP]

“We are leading efforts to establish international standards for Socamm2 in cooperation with major global partners,” said Samsung Electronics.

“Through ongoing [...] collaboration with Samsung Electronics, we are continuing to optimize next-generation memory such as Socamm2,” said Dion Harris, a senior director of high-performance computing and AI infrastructure solutions at Nvidia.

All three major memory chipmakers are competing to supply Socamm2 to Nvidia. In October, U.S.-based Micron, the world’s third-largest memory chipmaker, officially announced that it had supplied Socamm2 samples to Nvidia. On the same day, Samsung Electronics and SK hynix unveiled physical samples and specifications of their own Socamm2 at a semiconductor exhibition in Korea.

This article was originally written in Korean and translated by a bilingual reporter with the help of generative AI tools. It was then edited by a native English-speaking editor. All AI-assisted translations are reviewed and refined by our newsroom.
BY SHIM SEO-HYUN [[email protected]]