Samsung Electronics’ fourth-generation high-bandwidth memory (HBM3) chips have been approved by Nvidia for use in its processors for the first time, Reuters reported on July 24.
According to Reuters, citing three anonymous sources, Samsung’s HBM3 chips will only be used in a less advanced Nvidia graphics processing unit (GPU), the H20, developed for the Chinese market to comply with U.S. export controls, for now. The sources said it is unclear if Nvidia will use Samsung’s HBM3 chips in its other AI processors or if further tests are required. They added that Samsung could begin supplying HBM3 chips for Nvidia’s H20 processor as early as August.
The H20 has about one-fifth the computing power of Nvidia’s flagship AI chip H100 and has been supplied to the Chinese market since the first quarter of this year.
Since last year, Samsung has been undergoing Nvidia’s testing for both HBM3 and HBM3E. Currently, the fourth-generation HBM3 is the mainstream memory paired with Nvidia GPUs. SK Hynix, the exclusive supplier of HBM3 to Nvidia, began shipping HBM3E 8-layer products in March and aims to complete certification for 12-layer products by September.
Reuters also reported that Samsung’s fifth-generation HBM3E chips have not yet met Nvidia’s standards, and testing is still ongoing.
In response to the Reuters article, Samsung stated, “It is not true that HBM3E did not meet standards; the testing for HBM supply is progressing smoothly.”
