AI Infra Summit: NVIDIA Vera Rubin and DSX Platform Advancements Showcase Energy Efficiencies of Optimizing Tokens Per Watt for AI Factories
nvidia ·
At the AI Infra Summit, NVIDIA highlighted advancements in energy efficiency for AI factories, shifting the industry metric from peak performance to agentic tokens per megawatt. Ian Buck announced collaborations with partners like Amazon’s Annapurna Labs and d-Matrix to integrate custom memory and NVLink Fusion technologies. Lambda demonstrated a 23% improvement in performance per watt using NVIDIA DSX MaxLPS, allowing more nodes within the same power budget. Emerald AI showcased a flexible-load program with Silicon Valley Power, enabling AI facilities to dynamically adjust energy consumption based on grid signals. The new Vera Rubin NVL72 platform delivers up to 30x higher throughput per megawatt compared to previous generations, significantly reducing costs for agentic workloads. Additionally, startups reported substantial gains using the NVIDIA Vera CPU for secure sandboxes and analytical databases. These developments underscore NVIDIA's focus on codesigning silicon to grid to maximize productivity and economic value.
nvidia