Samsung Electronics has managed to reach an agreement with Nvidia on the supply of HBM3 memory chips. The “green” corporation will use them in some models of computing accelerators. In particular, the H20 AI chips with the Hopper architecture. This is a slightly cut-down version of the H100, created specifically for the Chinese market, taking into account the export restrictions of the United States.
The possibility of Samsung HBM3 chips appearing in top-end Nvidia accelerator models remains uncertain. According to Reuters, Samsung’s high-speed HBM3E chips are still undergoing testing in the “green” company’s labs. Heating and power consumption issues were reported a couple of months ago, while the South Korean giant itself denies this information.
Judging by the latest information, cooperation with Nvidia has become one of the main priorities for Samsung. In May, the Koreans allegedly implemented an internal strategy codenamed Nemo, aimed at priority fulfillment of Nvidia orders. All this should help Samsung get ahead of its main competitor, SK Hynix, which is currently the main supplier of HBM memory for Nvidia GPUs. Micron should also soon start supplying HBM3E chips.
Sources:
Reuters
Data Center Dynamics
The post Samsung to Supply HBM3 Chips for Some Nvidia AI Accelerators appeared first on Aroged.