NVIDIA Introduces Rubin CPX: Revolutionizing AI Processing
News Summary:
- The NVIDIA Rubin CPX GPU is purpose-built for complex applications like million-token coding and generative video.
- The Vera Rubin NVL144 CPX platform features 8 exaflops of AI capability and boasts 100TB of rapid memory.
- Businesses can achieve enormous profits, with $5 billion in token revenue generated for every $100 million invested.
- Companies such as Cursor, Runway, and Magic are eager to harness Rubin CPX to enhance their AI applications.
NVIDIA recently made waves by unveiling the Rubin CPX GPU, a cutting-edge component engineered for extensive context processing. Built to manage heavy workloads, the Rubin CPX will drastically improve the performance and efficiency of AI systems, particularly in cases like coding and video generation where processing millions of tokens is paramount.
The Rubin CPX operates synergistically with NVIDIA's Vera CPUs and Rubin GPUs within the innovative Vera Rubin NVL144 CPX platform. This powerful system delivers an impressive 8 exaflops of AI computational power, providing 7.5 times the operational capacity compared to previous systems. With 100TB of high-speed memory and a staggering memory bandwidth reaching 1.7 petabytes per second, this hardware setup stands to transform AI workloads significantly. For clients currently utilizing Vera Rubin 144 systems, a dedicated Rubin CPX compute tray will also be available.
Jensen Huang, NVIDIA's founder and CEO, shared his excitement over the launch, stating, "The Vera Rubin platform signifies a pivotal evolution in AI processing — presenting not only the revolutionary Rubin GPU but also introducing a new variety of processors known as CPX. Just as RTX transformed graphics and immersive AI, Rubin CPX is prime for scaling massive-context AI, enabling models to compare across millions of tokens simultaneously."
The capabilities of Rubin CPX, especially in long-context processing, are unprecedented. The GPU allows for enhanced performance and revenue potential, moving beyond primitive systems designed for simpler tasks. This innovation steps up AI coding assistants, transforming them from basic tools into advanced systems capable of navigating and optimizing extensive software projects.
For video applications, AI models can harness up to 1 million tokens for processing an hour's worth of footage, pushing the boundaries of traditional GPU capabilities. By consolidating video encoding and decoding into a single chip alongside long-context inference, Rubin CPX empowers impressive advancements in applications such as video analysis and high-quality generative content creation.
The Rubin architecture supports the Rubin CPX GPU with a cost-effective, monolithic die design, integrating potent NVFP4 computing elements to maximize performance while maintaining energy efficiency during inference tasks.
Key Features of Rubin CPX
Rubin CPX showcases the ability to deliver up to 30 petaflops of compute using NVFP4 precision, setting a new standard for performance and accuracy in calculation. It incorporates 128GB of efficient GDDR7 memory, elevating its capacity to tackle the most challenging workload contexts. Additionally, it boasts three times the attention processing speed compared to its predecessors, thus enhancing an AI model’s capability to handle longer context sequences without sacrificing performance.
This innovative GPU is available in several configurations, such as the Vera Rubin NVL144 CPX, designed to complement the NVIDIA Quantum-X800 InfiniBand scale-out compute fabric or the Spectrum-X™ Ethernet networking platform, which includes advanced technology like NVIDIA ConnectX®-9 SuperNICs. The Vera Rubin NVL144 CPX allows organizations to unlock remarkable profit margins, targeting $5 billion in token revenue for every $100 million invested.
Industry Adoption of Rubin CPX
Leading technology firms are eagerly exploring how Rubin CPX can drive innovation across various applications, from large-scale software development to dynamic visual content analysis.
For instance, Cursor, a software company equipped with an AI-driven code editor, is set to leverage Rubin CPX's capabilities to enhance developer productivity through intelligent code generation and real-time collaborative tools.
CEO of Cursor, Michael Truell, remarked, "With NVIDIA Rubin CPX, we expect to deliver ultra-fast code generation and valuable insights to developers, revolutionizing how software is built. This leap enables us to enhance productivity and facilitate the realization of previously unattainable ideas."
Runway, a leader in the generative AI sector, sees Rubin CPX as a game-changer for creatives. CEO Cristóbal Valenzuela stated, "As video production trends toward longer context and versatile creative tools, Rubin CPX represents significant performance advancements, empowering creators — from independent artists to major studios — with unprecedented speed and quality in their projects."
Magic, an AI research and product development firm, focuses on crafting foundational models to support AI that can autonomously manage software engineering tasks. According to CEO Eric Steinberger, "With a context window capable of hosting 100 million tokens, our models can analyze vast amounts of code and interaction history without continuous adjustment — facilitating direct training of AI agents through conversations and environmental access. Implementing a GPU like NVIDIA Rubin CPX greatly accelerates our operations."
Comprehensive Software Support
NVIDIA Rubin CPX will be integrated with a complete suite of NVIDIA AI tools, from high-performance infrastructure to enterprise-ready software. The NVIDIA Dynamo platform supports efficient AI inference and significantly enhances throughput while also reducing response times and operating costs.
The processors will run cutting-edge models from the NVIDIA Nemotron™ family, renowned for providing supreme reasoning capabilities for enterprise AI agents. Businesses can benefit from production-grade AI supported by NVIDIA AI Enterprise, which comes with essential microservices, toolkits, and frameworks for efficient deployment across various NVIDIA-accelerated environments.
Built on years of innovation, the Rubin platform is set to expand NVIDIA’s developer ecosystem, which already consists of a vast community comprising over 6 million developers and nearly 6,000 CUDA applications, supported by NVIDIA CUDA-X™ libraries.
Expected Availability
The NVIDIA Rubin CPX is projected to become accessible towards the close of 2026, marking a milestone in AI technology transitions.
If you're intrigued by these advancements, consider catching NVIDIA Vice President of Hyperscale and High-Performance Computing Ian Buck's keynote at the AI Infra Summit. It promises deeper insights into the transformative potential of the Rubin CPX GPU.
About NVIDIA
NVIDIA (NASDAQ: NVDA) stands as a global leader in accelerated computing, driving remarkable advancements across computing landscapes.
Contact for Further Information:
Kristin Uchiyama
NVIDIA Corporation
+1-408-313-0448
kuchiyama@nvidia.com
Frequently Asked Questions
What is the purpose of NVIDIA Rubin CPX?
The NVIDIA Rubin CPX GPU is designed to enhance massive-context processing for AI applications, particularly in coding and video generation.
How does Rubin CPX improve AI performance?
Rubin CPX provides up to 8 exaflops of AI computational power and optimizes memory bandwidth to handle extensive workloads efficiently.
What unique features does the Rubin CPX offer?
It offers advanced processing capabilities with NVFP4 precision, enhanced memory capacity, and unprecedented attention processing speeds.
Who are key innovators leveraging Rubin CPX?
Companies like Cursor, Runway, and Magic are utilizing Rubin CPX to boost their AI applications across various domains.
When will NVIDIA Rubin CPX be available?
The NVIDIA Rubin CPX is expected to be available towards the end of 2026, ushering in a new era of AI technology.
