Nvidia introduces Rubin CPX to enhance AI GPU capabilities

In early March 2025, Nvidia announced the introduction of the Rubin CPX, designed to work alongside its standard Rubin AI Graphics Processing Unit (GPU) to facilitate high-value inference content generation. The new accelerator emphasizes efficiency and cost-effectiveness in AI processing tasks, notably in industries requiring complex computations like software development and video generation. Following this announcement, the company launched its latest capabilities during the GTC 2025 event, focusing on improvements in AI-centric data center architectures including the NVL144 rack design and advanced networking technologies. The rationale behind the development of the Rubin CPX is rooted in the semiconductor industry's fast-paced technological advancements. Nvidia has recognized that GPUs, with their adaptability, serve as a more efficient solution compared to Application-Specific Integrated Circuits (ASICs) for various AI applications. This flexibility allows GPUs to accommodate rapid changes in AI models and processes. Furthermore, the architecture of GPUs enables concurrent handling of multiple AI models, making them indispensable in contemporary AI workloads, predominantly associated with generative and agentic AI. Nvidia is positioning the Rubin CPX within its broader ambition of creating comprehensive, self-sufficient AI data centers, or AI factories. This concept centers around building environments that optimize performance by addressing potential bottlenecks and maximizing return on investment (ROI). The Rubin CPX includes cutting-edge features such as 128GB of GDDR7 memory and specialized hardware for encoding and decoding video, making it particularly suitable for high-demand applications. It operates in tandem with the Vera CPU and Rubin AI GPU to create a synergistic and powerful processing unit. Additionally, the Vera Rubin NVL144 CPX configuration is expected to comprise a significant amount of processing power, including 36 Vera CPUs and 144 Rubin GPUs and CPXs, along with a staggering 100TB of high-speed memory. This transformation signifies Nvidia's commitment to evolving the data center landscape, with an eye towards scalability and long-term sustainability in AI research and commercial applications. As AI workloads evolve, the demand for innovative GPU solutions continues to surge, with Nvidia's initiatives positioning it at the forefront of this rapidly changing field.