This is not just some other new chip design — this is Nvidia’s Rubin architecture, a radical new chip design that will transform the future of high-performance computing. Still in production, Rubin is likely to increase its deployment significantly in the second half of 2024. This new architecture is dedicated to the esteemed astronomer Vera Florence Cooper Rubin. Inside, it utilizes six different chips working together to maximize computational efficiency while increasing speed.
Rubin architecture will be central to some of the most ambitious new supercomputing projects. This includes HPE’s Blue Lion supercomputer and the future Doudna supercomputer at Lawrence Berkeley National Lab. These systems will leverage the capabilities of the Rubin architecture to the fullest. They are purpose-built to handle the exponential workloads of specialized data processing, especially for AI applications.
At the heart of the Rubin architecture is our new Vera CPU, purpose-built for agentic reasoning. This step forward is a significant improvement compared to Nvidia’s previous Blackwell architecture. Blackwell itself is the model that replaced Hopper and Lovelace. Rubin is promising big leaps forward in speed and power consumption. It can deliver eight times more inference compute per watt than its predecessors.
Nvidia has entered into strategic partnerships with other major technology companies including Anthropic, OpenAI and Amazon Web Services. This partnership will help to ensure the widespread adoption of Rubin chips by the big cloud providers. We’re eager to see how this collaboration creates a new era of AI-assisted computational power. It addresses the growing storage and interconnection choke points by capitalizing on the latest breakthroughs in Bluefield and NVLink environments.
That’s why this innovation was emphasized as a need by Jensen Huang, CEO of Nvidia, saying it’s needed for the “impossible.”
“Vera Rubin is designed to address this fundamental challenge that we have: The amount of computation necessary for AI is skyrocketing.”
In Rubin architecture, computational capabilities are enhanced more than 30 times. It further introduces a new storage tier that’s externally connected to the compute device. Dion Harris, a principal architect of the technology’s development, said at the time,
“So we’ve introduced a new tier of storage that connects externally to the compute device, which allows you to scale your storage pool much more efficiently.”
The Rubin chips will be an incredible performance multiplier. Even better, they provide a highly scalable approach for building the new foundation of tomorrow’s distributed cloud computing and AI workloads. As production really starts to ramp up, there will be sea changes that shake up the industry. Supercomputers will do their work more seamlessly and analyze unprecedented amounts of data.