Blackwell shows how far NVIDIA has moved beyond graphics
NVIDIA's new architecture targets large AI workloads, making systems, interconnects and data-center design part of the chip story.
NVIDIA has announced Blackwell, its next major computing architecture, with a strong focus on the demands of generative AI. The announcement concerns more than a faster individual processor: NVIDIA is describing a platform intended to connect large amounts of compute into working systems.
Large models put pressure on memory movement and communication between chips as well as on arithmetic. That helps explain why the presentation extends to interconnects and complete system designs. For a data-center buyer, the surrounding infrastructure is part of the product decision.
Performance claims from a launch are only a beginning. Operators must translate them into their own workloads, power budgets and deployment plans. NVIDIA's broader message is clear, however: the company wants to supply the foundation of AI computing, not merely a component in somebody else's server.
Source: NVIDIA announcement ↗