Samsung Electronics’ fourth-generation high bandwidth memory (HBM3) chips have received approval from Nvidia for use in its processors, three sources familiar with the matter have revealed. However, this approval is limited to a less advanced Nvidia graphics processing unit (GPU), the H20, developed specifically for the Chinese market in compliance with U.S. export controls.
It remains unclear if Nvidia will utilize Samsung’s HBM3 chips in other AI processors or if further testing will be required, the sources added. Samsung has yet to meet Nvidia’s standards for its fifth-generation HBM3E chips, with testing still ongoing. Both Nvidia and Samsung have declined to comment on the matter.
HBM, a dynamic random access memory (DRAM) standard, features vertically stacked chips to save space and reduce power consumption. This technology, first produced in 2013, is essential for GPUs used in artificial intelligence, facilitating the processing of massive data sets from complex applications.
Nvidia’s approval of Samsung’s HBM3 chips comes amidst a surge in demand for advanced GPUs driven by the generative AI boom. This demand has outpaced supply, creating a pressing need for Nvidia to diversify its supplier base. Currently, there are only three main HBM manufacturers: SK Hynix, Micron, and Samsung.
Nvidia’s reliance on HBM3 is expected to grow as SK Hynix, the leader in the field, plans to increase production of the more advanced HBM3E and reduce HBM3 output, according to sources. SK Hynix, like Samsung, has been a key supplier of HBM chips to Nvidia, providing HBM3 since June 2022 and beginning HBM3E shipments in late March this year.
Samsung, the world’s largest memory chip maker, has been attempting to pass Nvidia’s tests for both HBM3 and HBM3E since last year but has faced challenges related to heat and power consumption. Despite a Reuters report in May citing these issues, Samsung refuted the claims, asserting that it had not failed Nvidia’s tests.
Sources indicate that Samsung could start supplying HBM3 for Nvidia’s H20 processor as early as August. The H20, designed for the Chinese market following tightened U.S. export restrictions in 2023, has significantly capped computing power compared to the H100, its counterpart for non-China markets. Initially priced below a rival chip from Huawei, the H20’s sales have been growing rapidly after a weak start.
Micron has also announced plans to supply Nvidia with HBM3E, joining SK Hynix and Samsung in supporting Nvidia’s expanding GPU demands. As Nvidia navigates the complexities of international regulations and market needs, securing reliable HBM suppliers remains crucial for its continued growth in AI technology.