Nvidia7 min readintermediate
AI Infra Summit: NVIDIA Vera Rubin and DSX Platform Advancements Showcase Energy Efficiencies of Optimizing Tokens Per Watt for AI Factories
Summary
NVIDIA showcased new power‑aware AI‑factory features at the AI Infra Summit. DSX MaxLPS and Flex improve token‑per‑watt efficiency by up to 40%, while the Vera Rubin NVL72 platform delivers up to 30× higher throughput per megawatt on agentic workloads.
- DSX MaxLPS lets 19 nodes run in the power budget of 16 full‑power nodes, boosting token throughput 24% and performance per watt 23%.
- Vera Rubin NVL72 with DSX MaxLPS can fit up to 40% more GPUs in the same site‑power envelope, delivering up to 35% higher token throughput.
- SemiAnalysis AgentX benchmark shows Vera Rubin NVL72 achieving up to 30× higher throughput per megawatt and 45× lower cost per million tokens for agentic workloads.
- DSX Flex enables AI factories to automatically throttle low‑priority jobs in response to grid demand‑response signals, turning data centers into flexible load resources.
AI infrastructure engineers and data‑center operators should care because these efficiency gains directly reduce energy costs and enable larger LLM deployments within existing power limits.
6/10
