Samsung Electronics has equipped next-generation DRAM module specification SOCAMM 2nd era samples to NVIDIA. This product is aimed toward being put in in NVIDIA’s next-generation synthetic intelligence (AI) chip Vera Rubin. The corporate plans to determine a mass manufacturing system in a well timed method to take care of early dominance within the SOCAMM market, which has emerged because the “second HBM.”
SOCAMM stands for small define copression hooked up reminiscence module and is a compact, high-performance reminiscence module optimized for AI servers. The module combines the high-bandwidth and vitality effectivity of LPDDR5X reminiscence with a modular, socketed kind issue.
Samsung Electronics introduced on Dec. 18 by means of its semiconductor newsroom that “we now have developed SOCAM2, an LPDDR-based server reminiscence module specialised for AI knowledge facilities, and are presently supplying samples to clients.”
The shopper talked about right here refers to NVIDIA. Dion Harris, NVIDIA’s basic supervisor of HPC and AI infrastructure options, stated, “By steady technical cooperation with Samsung Electronics, we’re persevering with optimization work in order that next-generation reminiscence resembling SOCAMM2 can implement the excessive responsiveness and effectivity required for AI infrastructure,” formally confirming the collaboration between the 2 firms.
SOCAMM is a detachable DRAM module primarily based on LPDDR DRAM, for which NVIDIA is independently selling requirements to achieve dominance within the AI reminiscence semiconductor market. Because it makes use of LPDDR5X DRAM, which has the benefit of low energy consumption, as an alternative of DDR-based server modules, it offers greater than twice the bandwidth and over 55% decrease energy consumption in comparison with current RDIMM. Velocity has additionally improved by greater than 20% in comparison with the earlier era SOCAMM1.
Samsung Electronics’ official announcement of SOCAMM pattern provision is interpreted as demonstrating its intention to enter the SOCAMM provide chain early. Samsung Electronics is understood to have entered the client pattern (CS) stage quicker than opponents by conducting shut cooperation with NVIDIA from the early growth stage. The CS stage is a key gateway for verifying stability and compatibility in precise system environments, that means it has met NVIDIA’s energy, bandwidth, and thermal administration requirements. If it proceeds easily to mass manufacturing, Samsung Electronics’ SOCAMM2 can be put in within the next-generation AI chip Vera Rubin.
With the SOCAMM2 market anticipated to bloom from the primary half of subsequent yr, reminiscence firms are engaged in fierce competitors to safe early dominance in NVIDIA’s provide chain. It’s because the market development fee of next-generation AI product traces is incomparably quick in comparison with earlier than, making it troublesome to catch the expansion pattern with out early response. Within the case of HBM, SK Hynix secured a dominant place early on, however for SOCAMM, alternatives to take the lead stay open. Samsung Electronics is evaluated as comparatively advantageous in competitors, having maintained the highest place in LPDDR product traces, which kind the idea of SOCAMM, for an extended interval.
Samsung Electronics’ technique is to take care of dominance in standardization primarily based on the benefit gained within the technique of getting into the SOCAMM provide chain. Presently, Samsung Electronics is main the institution of SOCAMM’s JEDEC normal specs along with main international companions.
A Samsung Electronics official defined, “We plan to repeatedly current options that present a balanced mixture of efficiency, energy, and scalability required by next-generation AI knowledge facilities by additional strengthening our server reminiscence product lineup.”
