HBM 4争夺战，正式打响

Core Viewpoint - NVIDIA's CEO Jensen Huang announced the full production of the next-generation AI chip platform "Vera Rubin," which is expected to intensify competition among major memory manufacturers, including Samsung, SK Hynix, and Micron [1] Group 1: Vera Rubin Production and Features - Vera Rubin has entered full-scale production and is set to be delivered to global customers by the end of this year, confirming earlier expectations of a mid-year release [1] - The platform integrates six chips, including Vera CPU and Rubin GPU, utilizing a strategy that connects multiple chips to enhance performance and efficiency, moving away from traditional methods of increasing transistor counts [1] Group 2: Performance Improvements - Compared to the previous Blackwell architecture, Vera Rubin improves AI inference performance by 5 times and training performance by 3.5 times, emphasizing the importance of system architecture in eliminating data bottlenecks [2] - Rubin GPU will feature HBM4 memory, which is anticipated to become mainstream in the HBM market following its release [2] Group 3: Memory Market Competition - The competition in the HBM4 and SOCAMM2 memory markets is expected to intensify, with Samsung planning to leverage its first 10nm-class HBM4 technology, achieving data processing speeds of 11Gbps and a bandwidth of 2.8TB/s, with a 40% improvement in energy efficiency over the previous generation [3] - SK Hynix currently holds a 57% market share in the global HBM market and plans to showcase its upgraded HBM4 products at CES, including SOCAMM2 solutions [3] Group 4: Industry Dynamics - Other companies, such as AMD, are also developing their own AI chips, indicating that competition among memory companies will significantly increase starting this year [4]