SK hynix is on track to have its cutting-edge HBM4 memory ready for NVIDIA’s next-generation Rubin R100 AI GPUs, slated for release in 2025. According to a recent report from ZDnet, SK hynix is in the final stages of commercializing its HBM4 memory, with the design drawings set to be transferred to the manufacturing process, known as ‘tape-out’, by October. This marks a crucial step towards bringing this high-performance memory to market.
HBM4 boasts a remarkable 40% increase in bandwidth compared to its predecessor, HBM3E, which is currently considered the fastest memory in the world. Furthermore, it achieves an impressive 70% reduction in power consumption, making it both powerful and efficient. HBM4 also features a 1.3x higher density, further enhancing its capabilities. This combination of advancements promises a significant leap in performance and efficiency, playing a key role in NVIDIA’s continued dominance in the AI GPU market.
NVIDIA has confirmed that its next-generation Rubin AI GPUs will utilize this advanced HBM4 memory. These GPUs are expected to be available in 2026, with a Rubin Ultra AI GPU offering following in 2027. The Rubin R100 AI GPUs are anticipated to use a 4x reticle design, an increase from the 3.3x reticle design used in the Blackwell generation. They are also set to be manufactured using TSMC’s advanced CoWoS-L packaging technology on the N3 process node.
TSMC has recently announced plans for chips with up to 5.5x reticle size, featuring a massive 100 x 100mm substrate capable of accommodating 12 HBM sites, compared to the current 8 HBM sites on 80 x 80mm packages. TSMC is also transitioning to a new SoIC design, which allows for larger reticle sizes, potentially reaching 8x or more on a 120 x 120mm package configuration. However, these plans are still in development, suggesting that the 4x reticle size for the Rubin R100 AI GPUs is a more realistic expectation for the near future.
The advancements in HBM4 memory and the development of NVIDIA’s next-generation Rubin AI GPUs are set to propel AI capabilities to new heights. With their impressive performance and efficiency gains, these technologies are poised to revolutionize various sectors, including research, development, and everyday applications.
