SK Hynix launched mass production of SOCAMM2 memory modules for Nvidia AI servers

South Korea’s SK Hynix has started mass production of next-generation memory modules for American Nvidia Corp.’s Vera Rubin platform, SK Hynix said in a press release.

The SOCAMM2 (Small Outline Compression Attached Memory Module 2) modules have a capacity of 192GB and are optimised for use in AI servers.

They have a slim form factor and are based on LPDDR5 memory, a low-power memory standard previously used primarily in mobile devices.

SK Hynix expects the SOCAMM2s to provide an opportunity to “fundamentally address” memory-related limitations in the context of training and operating AI models and thereby significantly improve the speed of the entire system. The company will work closely with Nvidia to achieve this.

The Vera Rubin platform combines the new Vera processors and Rubin GPUs (named after astrophysicist Vera Rubin).

- Реклама -