Nvidia finds Samsung's HBM4 chip to be the best in the market

Since GenAI chatbots hit the market, the demand for AI is skyrocketing. These chatbots run on servers with AI accelerators, which use high-bandwidth memory (HBM). As a result, the demand for AI accelerators and HBM is also increasing sharply. Currently, Nvidia is the most popular brand in the AI accelerator space, and its upcoming AI accelerator, Rubin, uses HBM4 memory.

At the moment, there are three brands ready to start the mass production of HBM4 chips: Micron, Samsung, and SK Hynix. Nvidia is expected to source HBM4 chips for Rubin from all three brands. To that extent, Nvidia has been testing HBM4 chips from the three companies. Well, it has found that Samsung’s HBM4 chips are the best among the lot, in terms of speed and power efficiency.

According to a new report from Maeil Kyungjae, Nvidia shared this news with Samsung personally during its visit to Samsung to discuss the progress of the HBM4 system-in-package (SiP). The publication says that Nvidia has passed Samsung’s HBM4 chip in the quality test.

The report also says that Nvidia requested Samsung to supply a significantly higher amount of HBM4 chips than Samsung internally projected, mostly because it performed the best. Samsung is expected to start supplying those modules to Nvidia in the first half of next year.

Samsung is reportedly running three months behind Micron and SK Hynix in terms of HBM4 development, production, and shipping. However, considering that it is the best among the lot, the delay seems to be worth it. 

Recently, Samsung reclaimed the second place in the HBM market from Micron. Looking at how the brand’s HBM4 chip is performing, the South Korean tech giant seems to have the potential to take the HBM market by storm.