NVIDIA has introduced the Rubin architecture, which is expected to radically change the field of AI. During a visit to Taiwan, company CEO Jensen Huang announced that six Rubin chips have already been sent to TSMC factories for production.
The new lineup includes not only CPUs and GPUs, but also an innovative NVLink switch, as well as a silicon-based photonic processor. Rubin will be the first architecture to support HBM4 memory, which will replace HBM3E, providing higher data transfer rates and reducing power consumption.
Production will be established using the 3 nm (N3P) process with advanced CoWoS-L packaging, which will allow handling requests in the field of AI. For the first time, NVIDIA will implement a chiplet architecture, combining different components into one solution.
Developers claim that Rubin will significantly surpass the capabilities of current Blackwell solutions. According to experts, the new product will be as significant as Hopper in 2022. The commercial release of Rubin is expected in 2026–2027, and the industry is already preparing for large-scale changes.
***
Cover author: LTN. Cover source: YouTube
