The New Engine of AI: NVIDIA Vera Rubin

As companies race to build ultimate AI, they all share a common challenge: the immense computational power required to train and simulate increasingly complex models. But what if there is a machine, that it's sole purpose is to handle this new era of intelligence? Named after the pioneering astronomer Vera Rubin, the NVIDIA Vera Rubin platform, the next generational architecture built to power the most demanding computing workload of the coming decades.

A Specialized System

The Vera Rubin platform is a sophisticated "alliance" of specialized chips working together. It includes:

• Vera CPUs: The next-generation central processing units designed by NVIDIA to work in perfect harmony with its GPUs.

• Rubin GPUs: The next-generation graphics processing units that are the primary workhorses for AI model training.

• Rubin CPX GPUs: A new class of GPUs specifically engineered for "massive-context Inference" These chips are designed to handle millions of tokens at once, making them ideal for complex tasks like understanding an entire software codebase or a full-length film.

From Grace to Vera: What are the differences?

The NVIDIA Vera Rubin platform represents a fundamental shift in architecture, not just a simple upgrade over its predecessor, the NVIDIA Grace Blackwell Superchip. While Grace Blackwell was a monumental achievement in its own right—revolutionizing large model training and inference—Vera Rubin is specifically engineered to solve the next big bottleneck in AI: long-context inference.

Compared to the Grace Blackwell platform, Vera Rubin's design is purpose-built to handle immense workloads that can involve millions of tokens at once—the equivalent of an entire software codebase or a full-length film. This is made possible through a new, specialized "alliance" of chips: the powerful Rubin GPUs and the brand-new, NVIDIA-designed Vera CPUs, which work together to boost performance.

This integrated approach allows the platform to achieve staggering performance. A single Vera Rubin system in a rack can deliver 8 exaflops of AI performance and a 100TB of fast memory with 1.7 petabytes per second of memory bandwidth, providing a 7.5x increase over the NVIDIA GB300 NVL (the Grace Blackwell Superchip).

Building the Road to the Future of Intelligence

The NVIDIA Vera Rubin platform is more than just the next step in a product roadmap; it is the new engine of an industrial revolution. As the demand for more intelligent, reasoning-based AI grows, Vera Rubin will provide the foundational computing power that enables new scientific discoveries and unlocks capabilities that were previously confined to science fiction. With its strategic alliances and specialized architecture, it stands as a testament to NVIDIA's vision to not only build the future of AI, but to be the force that powers it.