Nvidia unveils new GPU designed for long-context inference (techcrunch.com)

Nvidia has unveiled the Rubin CPX, a new GPU engineered specifically for handling extremely long context windows exceeding 1 million tokens. Announced at the AI Infrastructure Summit, this GPU is part of Nvidia’s upcoming Rubin series and is optimized to accelerate processing of large sequence data, making it ideal for applications like video generation and complex software development that demand extensive contextual understanding. This innovation marks a significant advancement for the AI and machine learning community by enabling more efficient inference on tasks with vast contextual requirements, which has traditionally been a major bottleneck. The Rubin CPX is designed to integrate within Nvidia’s broader “disaggregated inference” architecture, signaling a shift towards modular, scalable AI infrastructure that can tackle increasingly complex models without sacrificing performance. Slated for release at the end of 2026, the Rubin CPX underscores Nvidia’s leading role in AI hardware, supported by its robust data center revenue of $41.1 billion last quarter. This GPU is poised to empower next-generation AI workloads that rely heavily on long context processing, pushing the boundaries of what’s possible in large-scale, sequence-based AI applications.