NVIDIA CEO Jensen Huang takes center stage again at CES 2025, as he unveiled several AI advancements in addition to the powerful GeForce RTX 50 GPU, built on its next-gen Blackwell architecture. Notably, he also made an update on Samsung’s HBM3E verification, as per South Korean media outlet Korea JoongAng Daily.
Citing Huang’s remarks during the CES 2025 press Q&A session, the report notes that Samsung must develop a “new design” to pass NVIDIA’s qualification. “But they can do it, and they are working very fast,” Huang added.
Huang confirmed South Korea’s ambitions on HBM, noting that Samsung developed the first HBM NVIDIA ever used, and he believed the company would bounce back.
The report indicates that Huang has remained optimistic about Samsung’s HBM3E chip approval since 2024. In March 2024, he stated that Samsung’s HBM chips were in the “qualification” stage and praised the company as “extraordinary.” In November, he added that NVIDIA was working quickly to certify Samsung’s advanced AI memory chips.
According to TrendForce, Samsung’s recent struggles with HBM3E qualification indicate that 12hi solutions for Blackwell will require additional time for approval. TrendForce posits that while NVIDIA benefits from diversifying its HBM supplier base, it is critical that these suppliers provide reliable memory solutions to safeguard the costly chips integrated with HBM on an interposer.
It is worth noting that SK hynix, as NVIDIA’s primary HBM supplier, is already providing 8-layer and 12-layer HBM3E chips to NVIDIA and unveiled the world’s first 16-layer HBM3E chips at CES 2025. As per the Chosun Daily, SK hynix intends to provide NVIDIA with samples of its 16-layer HBM3E products in the first half of 2025 to begin the certification process.
RTX50 GPU Adopts Micron’s GDDR7 Chips
On the other hand, Huang revealed at CES that the newly-released RTX50 GPU is equipped with GDDR7 memory from U.S.-based Micron, as per another report from ZDNet.
The ZDNet report suggests that Huang’s decision to name Micron as a memory partner at a major event is seen as a gesture aimed at maintaining positive relations with domestic firms ahead of the second-term inauguration of U.S. President Trump on January 20.
Interestingly enough, the Korea JoongAng Daily report notes that when questioned about NVIDIA’s choice to use Micron’s GDDR7 instead of chips from Samsung or SK hynix, Huang replied, “I’m not exactly sure why. It’s probably nothing significant.”
Read more
(Photo credit: NVIDIA)