CEO of NVIDIA Jensen Huang announced that Samsung’s High Bandwidth Memory (HBM) did not fail the quality test. According to the industry, on June 4, Huang stated at a press conference at the Grand Hyatt Taipei in Taiwan, “We are collaborating with Samsung Electronics, SK Hynix, and Micron. We will receive products from all of them.” He added, “The companies are qualified, and we are working to apply them to our manufacturing system as quickly as possible.”
Regarding the recent reports that Samsung’s HBM failed NVIDIA’s quality test due to problems such as heat generation, he countered, “The test was ongoing even yesterday, and it’s not that it failed the test.” He further stated, “Our work with Samsung is going well. Patience is required.”
Reuters reported on May 23 that “the results of the failure of Samsung’s 8-layer and 12-layer HBM3E verification came out last month.”
Samsung currently focuses on passing the NVIDIA quality test of the 5th generation HBM, the 12-layer HBM3E.
Announcing the next-generation artificial intelligence (AI) graphics processing unit (GPU) called Rubin ahead of the opening of Computex 2024 in Taipei on June 2, Huang revealed that they plan to start mass production in 2026.
The media analyzed the unveiling of Rubin just three months after the reveal of Blackwell as a sign that NVIDIA is accelerating the development of AI semiconductors.
Rubin is scheduled to be delivered to customers in the second half of this year.
Huang announced that NVIDIA will announce new AI semiconductors annually, instead of every two years as before.
Releasing in 2027, it has been reported that NVIDIA plans to equip the Rubin GPU with eight HBM4s, the 6th generation HBM product, and the Rubin Ultra GPU with twelve HBM4s.
