Main South Korean reminiscence producer SK hynix announced that it’s going to showcase a collection of superior reminiscence options tailor-made for synthetic intelligence (AI) functions at this 12 months’s Consumer Electronics Show (CES) in Las Vegas.
Constructing upon its 12-layer Excessive Bandwidth Reminiscence (HBM) know-how, the corporate will show samples of its newest 16-layer HBM3E merchandise, formally announced in November 2024. This development employs superior MR-MUF processes to reinforce thermal efficiency and mitigate chip warping, reaching industry-leading outcomes.
With capacities of 48GB (3GB per particular person die) per stack, the elevated density will permit AI accelerators to make the most of as much as 384GB of HBM3E reminiscence in an 8-stack configuration. The 16-layer HBM3E is designed to considerably increase AI studying by as much as 18% and inference efficiency by as much as 32% in comparison with the 12-layer model.
Nvidia’s next-gen Rubin chips are slated for mass manufacturing later subsequent 12 months, thus the existence of HBM3E could possibly be shortlived, as the brand new upcoming Nvidia chips might be primarily based on HBM4. That should not be a priority, although, as reports indicate that SK hynix achieved its tape-out section in October 2024.
Addressing the escalating demand for high-capacity storage in AI information facilities, SK hynix may also introduce new SSD options for enterprise customers, together with the 122TB ‘D5-P5336’ enterprise SSD, developed by its subsidiary Solidigm. This mannequin is claimed to boast the best capability presently out there in its class and is poised to set new requirements in information storage options.
The reminiscence and storage producer may also speak about Compute Specific Hyperlink (CXL) and Processing-In-Reminiscence (PIM) applied sciences, that are stated to be pivotal to the subsequent era of knowledge heart infrastructures. Modularized options just like the CMM-Ax and AiMX might be featured, with the CMM-Ax being hailed as a groundbreaking resolution that mixes the scalability of CXL with computational capabilities, boosting efficiency and power effectivity for next-generation server platforms.
With on-device AI turning into a well-liked development, SK hynix additionally has plans to showcase ‘LPCAMM23’ and ‘ZUFS 4.04,’ designed to reinforce information processing velocity and energy effectivity in edge units comparable to PCs and smartphones. These improvements goal to facilitate the mixing of AI capabilities immediately into client electronics, broadening the scope of AI functions.
The corporate introduced final 12 months that it was additionally engaged on a variety of different merchandise, together with PCIe 6.0 SSDs, high-capacity QLC (Quad Degree Cell) eSSDs made particularly for AI servers, and UFS 5.0 for cellular units. SK hynix can be engaged on an LPCAMM2 module and soldered LPDDR5/6 reminiscence utilizing its 1cnm-node to energy laptops and handheld consoles.