NVIDIA (Nvidia), a major AI chip manufacturer, announced on the 9th that it will launch the new AI chip Rubin CPX built with the Rubin architecture at the end of 2026. It is designed for processing ultra-long contextual reasoning tasks and is suitable for complex applications such as video production and software development.
NVIDIA officially launched the new picture processor (GPU) called Rubin CPX, which belongs to the new NVIDIA chip architecture Rubin series, allowing AI models to operate faster and more efficiently when facing long sequence contexts.
NVIDIA executive chief Huang Ren came to the station last month. As soon as he left the airport, he talked about the new generation of chip architecture Rubin. He also said that NVIDIA has 6 new chips that cooperate with the TC, including new CPUs (central processors), GPUs, NVLink interchanger chips, Internet chips, silicon photonic chips, etc.
NVIDIA released the Rubin architecture last year, named after Vera Rubin, a female astronomer who discovered dark matter.
NVIDIA today pointed out that modern AI models must be able to handle complex tasks in the fields of software development, video generation and in-depth research. These tasks bring new challenges to computing and memory, and need to rethink how to expand and optimize AI reasoning.
In these challenges, specific types of work, such as handling a large number of contexts, are becoming increasingly critical.
NVIDIA said that Rubin CPX is a GPU designed to accelerate long context processing. It works with the NVIDIARubin GPU and Vera CPU to form a complete system to optimize the processing of complex AI tasks.
AI is developing rapidly, NVIDIA continues to accelerate its R&D cycle, with sales of data centers reaching US$41.1 billion (about NT$1249 billion). Rubin CPX is scheduled to be launched by the end of 2026.