← Back
NVIDIA Blog

AI Infra Summit: NVIDIA Vera Rubin and DSX Platform Advancements Showcase Energy Efficiencies of Optimizing Tokens Per Watt for AI Factories

•10 min read•
#compute#inference#nvidia
Level:Advanced
For:AI Engineers
✦TL;DR

NVIDIA showcased its Vera Rubin and DSX platforms at the AI Infra Summit, highlighting a 1.4× boost in tokens per megawatt with DSX MaxLPS and a 3.7× higher throughput for the Vera Rubin NVL72 system over the prior GB300 NVL72. The announcements emphasized new collaborations (e.g., Amazon Annapurna Labs on NVHBM, d‑Matrix with NVLink Fusion) and a full‑stack AI factory that spans CPUs, XPUs, networking, and software to maximize token generation efficiency. For engineers, these advances translate into measurable power‑efficiency gains and dramatically higher inference throughput on a unified NVIDIA infrastructure.

⚡ Key Takeaways

  • DSX MaxLPS delivers up to 1.4× more tokens per megawatt through factory‑wide power optimization.
  • NVLink Fusion integrates Vera CPUs with d‑Matrix Raptor XPUs for ultra‑low‑latency inference at scale.
  • The Vera Rubin NVL72 system achieved up to 3.7× higher MLPerf Inference v6.1 throughput compared to the GB300 NVL72.
  • Dynamo inference software and NeMo libraries are part of the full‑stack platform for flexible model deployment.
  • Integration requires NVIDIA networking components (NVLink, Spectrum‑X Ethernet, ConnectX SuperNICs, BlueField DPUs) to connect thousands of nodes.
💡 Why It Matters

Engineers can now design AI factories that generate significantly more tokens per unit power, reducing operational costs while scaling inference workloads across heterogeneous hardware.

✅ Practical Steps

  1. Deploy NVIDIA DSX MaxLPS on your AI factory to achieve the advertised 1.4× token‑per‑megawatt improvement.
  2. Integrate NVLink Fusion between Vera CPUs and d‑Matrix Raptor XPUs to enable ultra‑low‑latency inference at scale.
  3. Use Dynamo inference software together with NeMo libraries on the Vera Rubin platform to run any model across training, inference, recommender, reasoning, and generative workloads.

Want the full story? Read the original article.

Read on NVIDIA Blog ↗

More like this

AI’s GPU focus meets a CPU reality check at Dell and AMD

SiliconANGLE AI•#agents

NVIDIA RTX Spark Compact Desktops Out Now

NVIDIA Blog•#compute

Helping AI text detectors keep up with changing conditions

MIT News AI•#llm

Hugging Face and OpenRouter join forces to make open-weight models easy to run

Hugging Face Blog•#inference

EXPLORE AI NEWS

Daily hand-picked stories on LLMs, RAG, agents and production AI — curated for engineers who ship.

BROWSE NEWS

GET THE WEEKLY DIGEST

Join engineers getting the Monday signal-over-noise AI breakdown. No spam, unsubscribe anytime.

LEARN AI ENGINEERING

Curated courses, research papers, repos and tutorials built for engineers leveling up in AI.

START LEARNING