According to reports from South Korea, NVIDIA’s key memory supplier, SK hynix, may begin early production of HBM4 memory, a critical component for NVIDIA’s upcoming Rubin GPU architecture. This development could lead to the launch of Rubin as early as the second half of 2025, six months ahead of the initial schedule.
Named after astronomer Vera Rubin, this architecture follows the recently released Blackwell series, which powers the GeForce RTX 5090, RTX 5080, RTX 5070 Ti, and RTX 5070 GPUs. Despite its recent launch, NVIDIA appears determined to maintain a fast innovation pace to meet the rising demand for AI computational power.
HBM4 Memory
HBM4, a high-bandwidth memory technology, is central to Rubin’s capabilities. It integrates memory and logic into a single package, boosting performance and energy efficiency while reducing packaging complexity. With data transfer speeds reaching up to 6.4 Gbps, HBM4 represents a significant leap forward, enabling more advanced and powerful AI systems.
Rubin will also leverage TSMC’s 3nm manufacturing process and CoWoS-L packaging technology. NVIDIA aims to achieve exceptional performance per watt, addressing the energy demands of modern AI systems and promoting long-term sustainability without compromising power.
NVIDIA Strengthens Its Leadership in AI
Launching Rubin ahead of schedule would reinforce NVIDIA’s leadership in AI technologies. Competitors are expected to release alternatives several quarters later, giving NVIDIA a significant head start in the market.
