Nvidia unveils new GPU designed for long-context inference

During the AI Infrastructure Summit held on Tuesday, Nvidia introduced its latest GPU, the Rubin CPX, specifically engineered to handle context windows exceeding 1 million tokens. This new addition is part of the upcoming Rubin series and is meticulously optimized for processing extensive sequences of context. The Rubin CPX is designed to integrate seamlessly into a comprehensive ‘disaggregated inference’ infrastructure, promising users improved performance for a range of long-context applications, including video generation and software development. Nvidia's continuous innovation has led to remarkable financial success, with the company reporting $41.1 billion in data center revenue for the last quarter alone. Set to hit the market by the end of 2026, the Rubin CPX signifies another step forward in Nvidia's commitment to advancing AI capabilities, ensuring that users have the tools necessary for increasingly complex tasks.